d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
5dab8f30b6
freebsd-skq/sys/kern/kern_umtx.c

4713 lines
107 KiB
C
Raw Normal View History

/* -> /*- for copyright notices, minor format tweaks as necessary
2005-01-06 23:35:40 +00:00
/*-
sys/kern: adoption of SPDX licensing ID tags. Mainly focus on files that use BSD 2-Clause license, however the tool I was using misidentified many licenses so this was mostly a manual - error prone - task. The Software Package Data Exchange (SPDX) group provides a specification to make it easier for automated tools to detect and summarize well known opensource licenses. We are gradually adopting the specification, noting that the tags are considered only advisory and do not, in any way, superceed or replace the license texts.
2017-11-27 15:20:12 +00:00
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
* Copyright (c) 2015, 2016 The FreeBSD Foundation
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
* Copyright (c) 2004, David Xu <davidxu@freebsd.org>
- Add an api for doing smp safe locks in userland. - umtx_lock() is defined as an inline in umtx.h. It tries to do an uncontested acquire of a lock which falls back to the _umtx_lock() system-call if that fails. - umtx_unlock() is also an inline which falls back to _umtx_unlock() if the uncontested unlock fails. - Locks are keyed off of the thr_id_t of the currently running thread which is currently just the pointer to the 'struct thread' in kernel. - _umtx_lock() uses the proc pointer to synchronize access to blocked thread queues which are stored in the first blocked thread.
2003-04-01 01:10:42 +00:00
* Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org>
* All rights reserved.
*
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
* Portions of this software were developed by Konstantin Belousov
* under sponsorship from the FreeBSD Foundation.
*
- Add an api for doing smp safe locks in userland. - umtx_lock() is defined as an inline in umtx.h. It tries to do an uncontested acquire of a lock which falls back to the _umtx_lock() system-call if that fails. - umtx_unlock() is also an inline which falls back to _umtx_unlock() if the uncontested unlock fails. - Locks are keyed off of the thr_id_t of the currently running thread which is currently just the pointer to the 'struct thread' in kernel. - _umtx_lock() uses the proc pointer to synchronize access to blocked thread queues which are stored in the first blocked thread.
2003-04-01 01:10:42 +00:00
* 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 unmodified, 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR 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.
*/
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Add rudimentary profiling of the hash table used in the in the umtx code to hold active lock queues. Reviewed by: attilio Approved by: davidxu, gnn (mentor) MFC after: 3 weeks
2012-03-16 20:32:11 +00:00
#include "opt_umtx_profiling.h"
- Add an api for doing smp safe locks in userland. - umtx_lock() is defined as an inline in umtx.h. It tries to do an uncontested acquire of a lock which falls back to the _umtx_lock() system-call if that fails. - umtx_unlock() is also an inline which falls back to _umtx_unlock() if the uncontested unlock fails. - Locks are keyed off of the thr_id_t of the currently running thread which is currently just the pointer to the 'struct thread' in kernel. - _umtx_lock() uses the proc pointer to synchronize access to blocked thread queues which are stored in the first blocked thread.
2003-04-01 01:10:42 +00:00
#include <sys/param.h>
#include <sys/kernel.h>
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
#include <sys/fcntl.h>
#include <sys/file.h>
#include <sys/filedesc.h>
Change the thread ID (thr_id_t) used for 1:1 threading from being a pointer to the corresponding struct thread to the thread ID (lwpid_t) assigned to that thread. The primary reason for this change is that libthr now internally uses the same ID as the debugger and the kernel when referencing to a kernel thread. This allows us to implement the support for debugging without additional translations and/or mappings. To preserve the ABI, the 1:1 threading syscalls, including the umtx locking API have not been changed to work on a lwpid_t. Instead the 1:1 threading syscalls operate on long and the umtx locking API has not been changed except for the contested bit. Previously this was the least significant bit. Now it's the most significant bit. Since the contested bit should not be tested by userland, this change is not expected to be visible. Just to be sure, UMTX_CONTESTED has been removed from <sys/umtx.h>. Reviewed by: mtm@ ABI preservation tested on: i386, ia64
2004-07-02 00:40:07 +00:00
#include <sys/limits.h>
- Add an api for doing smp safe locks in userland. - umtx_lock() is defined as an inline in umtx.h. It tries to do an uncontested acquire of a lock which falls back to the _umtx_lock() system-call if that fails. - umtx_unlock() is also an inline which falls back to _umtx_unlock() if the uncontested unlock fails. - Locks are keyed off of the thr_id_t of the currently running thread which is currently just the pointer to the 'struct thread' in kernel. - _umtx_lock() uses the proc pointer to synchronize access to blocked thread queues which are stored in the first blocked thread.
2003-04-01 01:10:42 +00:00
#include <sys/lock.h>
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
#include <sys/malloc.h>
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
#include <sys/mman.h>
- Add an api for doing smp safe locks in userland. - umtx_lock() is defined as an inline in umtx.h. It tries to do an uncontested acquire of a lock which falls back to the _umtx_lock() system-call if that fails. - umtx_unlock() is also an inline which falls back to _umtx_unlock() if the uncontested unlock fails. - Locks are keyed off of the thr_id_t of the currently running thread which is currently just the pointer to the 'struct thread' in kernel. - _umtx_lock() uses the proc pointer to synchronize access to blocked thread queues which are stored in the first blocked thread.
2003-04-01 01:10:42 +00:00
#include <sys/mutex.h>
Sweep kernel replacing suser(9) calls with priv(9) calls, assigning specific privilege names to a broad range of privileges. These may require some future tweaking. Sponsored by: nCircle Network Security, Inc. Obtained from: TrustedBSD Project Discussed on: arch@ Reviewed (at least in part) by: mlaier, jmg, pjd, bde, ceri, Alex Lyashkov <umka at sevcity dot net>, Skip Ford <skip dot ford at verizon dot net>, Antoine Brodin <antoine dot brodin at laposte dot net>
2006-11-06 13:42:10 +00:00
#include <sys/priv.h>
- Add an api for doing smp safe locks in userland. - umtx_lock() is defined as an inline in umtx.h. It tries to do an uncontested acquire of a lock which falls back to the _umtx_lock() system-call if that fails. - umtx_unlock() is also an inline which falls back to _umtx_unlock() if the uncontested unlock fails. - Locks are keyed off of the thr_id_t of the currently running thread which is currently just the pointer to the 'struct thread' in kernel. - _umtx_lock() uses the proc pointer to synchronize access to blocked thread queues which are stored in the first blocked thread.
2003-04-01 01:10:42 +00:00
#include <sys/proc.h>
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
#include <sys/resource.h>
#include <sys/resourcevar.h>
#include <sys/rwlock.h>
Improve UMTX_PROFILING: - Use u_int values for length and max_length values - Add a way to reset the max_length heuristic in order to have the possibility to reuse the mechanism consecutively without rebooting the machine - Add a way to quick display top5 contented buckets in the system for the max_length value. This should give a quick overview on the quality of the hash table distribution. Sponsored by: EMC / Isilon storage division Reviewed by: jeff, davide
2013-03-09 15:31:19 +00:00
#include <sys/sbuf.h>
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
#include <sys/sched.h>
Add a lwpid field into per-cpu structure, the lwpid represents current running thread's id on each cpu. This allow us to add in-kernel adaptive spin for user level mutex. While spinning in user space is possible, without correct thread running state exported from kernel, it hardly can be implemented efficiently without wasting cpu cycles, however exporting thread running state unlikely will be implemented soon as it has to design and stablize interfaces. This implementation is transparent to user space, it can be disabled dynamically. With this change, mutex ping-pong program's performance is improved massively on SMP machine. performance of mysql super-smack select benchmark is increased about 7% on Intel dual dual-core2 Xeon machine, it indicates on systems which have bunch of cpus and system-call overhead is low (athlon64, opteron, and core-2 are known to be fast), the adaptive spin does help performance. Added sysctls: kern.threads.umtx_dflt_spins if the sysctl value is non-zero, a zero umutex.m_spincount will cause the sysctl value to be used a spin cycle count. kern.threads.umtx_max_spins the sysctl sets upper limit of spin cycle count. Tested on: Athlon64 X2 3800+, Dual Xeon 5130
2006-12-20 04:40:39 +00:00
#include <sys/smp.h>
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
#include <sys/sysctl.h>
- Add an api for doing smp safe locks in userland. - umtx_lock() is defined as an inline in umtx.h. It tries to do an uncontested acquire of a lock which falls back to the _umtx_lock() system-call if that fails. - umtx_unlock() is also an inline which falls back to _umtx_unlock() if the uncontested unlock fails. - Locks are keyed off of the thr_id_t of the currently running thread which is currently just the pointer to the 'struct thread' in kernel. - _umtx_lock() uses the proc pointer to synchronize access to blocked thread queues which are stored in the first blocked thread.
2003-04-01 01:10:42 +00:00
#include <sys/sysent.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
#include <sys/syscallsubr.h>
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
#include <sys/taskqueue.h>
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
#include <sys/time.h>
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
#include <sys/eventhandler.h>
- Add an api for doing smp safe locks in userland. - umtx_lock() is defined as an inline in umtx.h. It tries to do an uncontested acquire of a lock which falls back to the _umtx_lock() system-call if that fails. - umtx_unlock() is also an inline which falls back to _umtx_unlock() if the uncontested unlock fails. - Locks are keyed off of the thr_id_t of the currently running thread which is currently just the pointer to the 'struct thread' in kernel. - _umtx_lock() uses the proc pointer to synchronize access to blocked thread queues which are stored in the first blocked thread.
2003-04-01 01:10:42 +00:00
#include <sys/umtx.h>
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
#include <security/mac/mac_framework.h>
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_object.h>
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
#include <machine/atomic.h>
Add a lwpid field into per-cpu structure, the lwpid represents current running thread's id on each cpu. This allow us to add in-kernel adaptive spin for user level mutex. While spinning in user space is possible, without correct thread running state exported from kernel, it hardly can be implemented efficiently without wasting cpu cycles, however exporting thread running state unlikely will be implemented soon as it has to design and stablize interfaces. This implementation is transparent to user space, it can be disabled dynamically. With this change, mutex ping-pong program's performance is improved massively on SMP machine. performance of mysql super-smack select benchmark is increased about 7% on Intel dual dual-core2 Xeon machine, it indicates on systems which have bunch of cpus and system-call overhead is low (athlon64, opteron, and core-2 are known to be fast), the adaptive spin does help performance. Added sysctls: kern.threads.umtx_dflt_spins if the sysctl value is non-zero, a zero umutex.m_spincount will cause the sysctl value to be used a spin cycle count. kern.threads.umtx_max_spins the sysctl sets upper limit of spin cycle count. Tested on: Athlon64 X2 3800+, Dual Xeon 5130
2006-12-20 04:40:39 +00:00
#include <machine/cpu.h>
Provide groundwork for 32-bit binary compatibility on non-x86 platforms, for upcoming 64-bit PowerPC and MIPS support. This renames the COMPAT_IA32 option to COMPAT_FREEBSD32, removes some IA32-specific code from MI parts of the kernel and enhances the freebsd32 compatibility code to support big-endian platforms. Reviewed by: kib, jhb
2010-03-11 14:49:06 +00:00
#ifdef COMPAT_FREEBSD32
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
#include <compat/freebsd32/freebsd32_proto.h>
#endif
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
#define _UMUTEX_TRY 1
#define _UMUTEX_WAIT 2
Improve UMTX_PROFILING: - Use u_int values for length and max_length values - Add a way to reset the max_length heuristic in order to have the possibility to reuse the mechanism consecutively without rebooting the machine - Add a way to quick display top5 contented buckets in the system for the max_length value. This should give a quick overview on the quality of the hash table distribution. Sponsored by: EMC / Isilon storage division Reviewed by: jeff, davide
2013-03-09 15:31:19 +00:00
#ifdef UMTX_PROFILING
#define UPROF_PERC_BIGGER(w, f, sw, sf) \
(((w) > (sw)) || ((w) == (sw) && (f) > (sf)))
#endif
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* Priority inheritance mutex info. */
struct umtx_pi {
/* Owner thread */
struct thread *pi_owner;
/* Reference count */
int pi_refcount;
/* List entry to link umtx holding by thread */
TAILQ_ENTRY(umtx_pi) pi_link;
/* List entry in hash */
TAILQ_ENTRY(umtx_pi) pi_hashlink;
/* List for waiters */
TAILQ_HEAD(,umtx_q) pi_blocked;
/* Identify a userland lock object */
struct umtx_key pi_key;
};
/* A userland synchronous object user. */
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
struct umtx_q {
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* Linked list for the hash. */
TAILQ_ENTRY(umtx_q) uq_link;
/* Umtx key. */
struct umtx_key uq_key;
/* Umtx flags. */
int uq_flags;
#define UQF_UMTXQ 0x0001
/* The thread waits on. */
struct thread *uq_thread;
/*
* Blocked on PI mutex. read can use chain lock
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
* or umtx_lock, write must have both chain lock and
* umtx_lock being hold.
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
*/
struct umtx_pi *uq_pi_blocked;
/* On blocked list */
TAILQ_ENTRY(umtx_q) uq_lockq;
/* Thread contending with us */
TAILQ_HEAD(,umtx_pi) uq_pi_contested;
Reorder some statments. Fix typo and remove stale comments.
2006-08-30 23:59:45 +00:00
/* Inherited priority from PP mutex */
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
u_char uq_inherited_pri;
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
/* Spare queue ready to be reused */
struct umtxq_queue *uq_spare_queue;
/* The queue we on */
struct umtxq_queue *uq_cur_queue;
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
};
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
TAILQ_HEAD(umtxq_head, umtx_q);
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
/* Per-key wait-queue */
struct umtxq_queue {
struct umtxq_head head;
struct umtx_key key;
LIST_ENTRY(umtxq_queue) link;
int length;
};
LIST_HEAD(umtxq_list, umtxq_queue);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* Userland lock object's wait-queue chain */
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
struct umtxq_chain {
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* Lock for this chain. */
struct mtx uc_lock;
/* List of sleep queues. */
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
struct umtxq_list uc_queue[2];
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
#define UMTX_SHARED_QUEUE 0
#define UMTX_EXCLUSIVE_QUEUE 1
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
LIST_HEAD(, umtxq_queue) uc_spare_queue;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* Busy flag */
char uc_busy;
/* Chain lock waiters */
Use wakeup_one to avoid thundering herd. Tested by: kris
2006-05-09 13:00:46 +00:00
int uc_waiters;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* All PI in the list */
TAILQ_HEAD(,umtx_pi) uc_pi_list;
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
Add rudimentary profiling of the hash table used in the in the umtx code to hold active lock queues. Reviewed by: attilio Approved by: davidxu, gnn (mentor) MFC after: 3 weeks
2012-03-16 20:32:11 +00:00
#ifdef UMTX_PROFILING
Improve UMTX_PROFILING: - Use u_int values for length and max_length values - Add a way to reset the max_length heuristic in order to have the possibility to reuse the mechanism consecutively without rebooting the machine - Add a way to quick display top5 contented buckets in the system for the max_length value. This should give a quick overview on the quality of the hash table distribution. Sponsored by: EMC / Isilon storage division Reviewed by: jeff, davide
2013-03-09 15:31:19 +00:00
u_int length;
u_int max_length;
Add rudimentary profiling of the hash table used in the in the umtx code to hold active lock queues. Reviewed by: attilio Approved by: davidxu, gnn (mentor) MFC after: 3 weeks
2012-03-16 20:32:11 +00:00
#endif
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
};
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
#define UMTXQ_LOCKED_ASSERT(uc) mtx_assert(&(uc)->uc_lock, MA_OWNED)
/*
* Don't propagate time-sharing priority, there is a security reason,
* a user can simply introduce PI-mutex, let thread A lock the mutex,
* and let another thread B block on the mutex, because B is
* sleeping, its priority will be boosted, this causes A's priority to
* be boosted via priority propagating too and will never be lowered even
* if it is using 100%CPU, this is unfair to other processes.
*/
Make KSE a kernel option, turned on by default in all GENERIC kernel configs except sun4v (which doesn't process signals properly with KSE). Reviewed by: davidxu@
2006-10-26 21:42:22 +00:00
#define UPRI(td) (((td)->td_user_pri >= PRI_MIN_TIMESHARE &&\
(td)->td_user_pri <= PRI_MAX_TIMESHARE) ?\
PRI_MAX_TIMESHARE : (td)->td_user_pri)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
#define GOLDEN_RATIO_PRIME 2654404609U
[mips] make UMTX_CHAINS configurable at compile time. The default (512) wastes quite a bit of space which doesn't really buy us much on highly embedded systems which don't take a lot of locks in parallel. This makes it at least build time configurable so people can experiment.
2016-11-15 01:34:38 +00:00
#ifndef UMTX_CHAINS
Enlarge hash table for new condition variable.
2010-12-23 03:12:03 +00:00
#define UMTX_CHAINS 512
[mips] make UMTX_CHAINS configurable at compile time. The default (512) wastes quite a bit of space which doesn't really buy us much on highly embedded systems which don't take a lot of locks in parallel. This makes it at least build time configurable so people can experiment.
2016-11-15 01:34:38 +00:00
#endif
Enlarge hash table for new condition variable.
2010-12-23 03:12:03 +00:00
#define UMTX_SHIFTS (__WORD_BIT - 9)
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
#define GET_SHARE(flags) \
(((flags) & USYNC_PROCESS_SHARED) == 0 ? THREAD_SHARE : PROCESS_SHARE)
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
#define BUSY_SPINS 200
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout {
int clockid;
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
bool is_abs_real; /* TIMER_ABSTIME && CLOCK_REALTIME* */
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct timespec cur;
struct timespec end;
};
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
#ifdef COMPAT_FREEBSD32
struct umutex32 {
volatile __lwpid_t m_owner; /* Owner of the mutex */
__uint32_t m_flags; /* Flags of the mutex */
__uint32_t m_ceilings[2]; /* Priority protect ceiling */
__uint32_t m_rb_lnk; /* Robust linkage */
__uint32_t m_pad;
__uint32_t m_spare[2];
};
_Static_assert(sizeof(struct umutex) == sizeof(struct umutex32), "umutex32");
_Static_assert(__offsetof(struct umutex, m_spare[0]) ==
__offsetof(struct umutex32, m_spare[0]), "m_spare32");
#endif
int umtx_shm_vnobj_persistent = 0;
SYSCTL_INT(_kern_ipc, OID_AUTO, umtx_vnode_persistent, CTLFLAG_RWTUN,
&umtx_shm_vnobj_persistent, 0,
"False forces destruction of umtx attached to file, on last close");
static int umtx_max_rb = 1000;
SYSCTL_INT(_kern_ipc, OID_AUTO, umtx_max_robust, CTLFLAG_RWTUN,
&umtx_max_rb, 0,
"");
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static uma_zone_t umtx_pi_zone;
Use a seperated hash table for mutex and rwlock, avoid wasting some time on walking through idle threads sleeping on condition variables.
2008-05-30 02:18:54 +00:00
static struct umtxq_chain umtxq_chains[2][UMTX_CHAINS];
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
static MALLOC_DEFINE(M_UMTX, "umtx", "UMTX queue memory");
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static int umtx_pi_allocated;
- Create a new lock, umtx_lock, for use instead of the proc lock for protecting the umtx queues. We can't use the proc lock because we need to hold the lock across calls to casuptr, which can fault. Approved by: re
2003-05-25 18:18:32 +00:00
Mark all SYSCTL_NODEs static that have no corresponding SYSCTL_DECLs. The SYSCTL_NODE macro defines a list that stores all child-elements of that node. If there's no SYSCTL_DECL macro anywhere else, there's no reason why it shouldn't be static.
2011-11-07 15:43:11 +00:00
static SYSCTL_NODE(_debug, OID_AUTO, umtx, CTLFLAG_RW, 0, "umtx debug");
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
SYSCTL_INT(_debug_umtx, OID_AUTO, umtx_pi_allocated, CTLFLAG_RD,
&umtx_pi_allocated, 0, "Allocated umtx_pi");
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
static int umtx_verbose_rb = 1;
SYSCTL_INT(_debug_umtx, OID_AUTO, robust_faults_verbose, CTLFLAG_RWTUN,
&umtx_verbose_rb, 0,
"");
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Add rudimentary profiling of the hash table used in the in the umtx code to hold active lock queues. Reviewed by: attilio Approved by: davidxu, gnn (mentor) MFC after: 3 weeks
2012-03-16 20:32:11 +00:00
#ifdef UMTX_PROFILING
static long max_length;
SYSCTL_LONG(_debug_umtx, OID_AUTO, max_length, CTLFLAG_RD, &max_length, 0, "max_length");
static SYSCTL_NODE(_debug_umtx, OID_AUTO, chains, CTLFLAG_RD, 0, "umtx chain stats");
#endif
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
static void abs_timeout_update(struct abs_timeout *timo);
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
static void umtx_shm_init(void);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static void umtxq_sysinit(void *);
static void umtxq_hash(struct umtx_key *key);
static struct umtxq_chain *umtxq_getchain(struct umtx_key *key);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
static void umtxq_lock(struct umtx_key *key);
static void umtxq_unlock(struct umtx_key *key);
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
static void umtxq_busy(struct umtx_key *key);
static void umtxq_unbusy(struct umtx_key *key);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
static void umtxq_insert_queue(struct umtx_q *uq, int q);
static void umtxq_remove_queue(struct umtx_q *uq, int q);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
static int umtxq_sleep(struct umtx_q *uq, const char *wmesg, struct abs_timeout *);
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
static int umtxq_count(struct umtx_key *key);
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
static struct umtx_pi *umtx_pi_alloc(int);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static void umtx_pi_free(struct umtx_pi *pi);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
static int do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags,
bool rb);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static void umtx_thread_cleanup(struct thread *td);
static void umtx_exec_hook(void *arg __unused, struct proc *p __unused,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
struct image_params *imgp __unused);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
SYSINIT(umtx, SI_SUB_EVENTHANDLER+1, SI_ORDER_MIDDLE, umtxq_sysinit, NULL);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
#define umtxq_signal(key, nwake) umtxq_signal_queue((key), (nwake), UMTX_SHARED_QUEUE)
#define umtxq_insert(uq) umtxq_insert_queue((uq), UMTX_SHARED_QUEUE)
#define umtxq_remove(uq) umtxq_remove_queue((uq), UMTX_SHARED_QUEUE)
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
static struct mtx umtx_lock;
Add rudimentary profiling of the hash table used in the in the umtx code to hold active lock queues. Reviewed by: attilio Approved by: davidxu, gnn (mentor) MFC after: 3 weeks
2012-03-16 20:32:11 +00:00
#ifdef UMTX_PROFILING
static void
umtx_init_profiling(void)
{
struct sysctl_oid *chain_oid;
char chain_name[10];
int i;
for (i = 0; i < UMTX_CHAINS; ++i) {
snprintf(chain_name, sizeof(chain_name), "%d", i);
chain_oid = SYSCTL_ADD_NODE(NULL,
SYSCTL_STATIC_CHILDREN(_debug_umtx_chains), OID_AUTO,
chain_name, CTLFLAG_RD, NULL, "umtx hash stats");
SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
"max_length0", CTLFLAG_RD, &umtxq_chains[0][i].max_length, 0, NULL);
SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
"max_length1", CTLFLAG_RD, &umtxq_chains[1][i].max_length, 0, NULL);
}
}
Improve UMTX_PROFILING: - Use u_int values for length and max_length values - Add a way to reset the max_length heuristic in order to have the possibility to reuse the mechanism consecutively without rebooting the machine - Add a way to quick display top5 contented buckets in the system for the max_length value. This should give a quick overview on the quality of the hash table distribution. Sponsored by: EMC / Isilon storage division Reviewed by: jeff, davide
2013-03-09 15:31:19 +00:00
static int
sysctl_debug_umtx_chains_peaks(SYSCTL_HANDLER_ARGS)
{
char buf[512];
struct sbuf sb;
struct umtxq_chain *uc;
u_int fract, i, j, tot, whole;
u_int sf0, sf1, sf2, sf3, sf4;
u_int si0, si1, si2, si3, si4;
u_int sw0, sw1, sw2, sw3, sw4;
sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
for (i = 0; i < 2; i++) {
tot = 0;
for (j = 0; j < UMTX_CHAINS; ++j) {
uc = &umtxq_chains[i][j];
mtx_lock(&uc->uc_lock);
tot += uc->max_length;
mtx_unlock(&uc->uc_lock);
}
if (tot == 0)
sbuf_printf(&sb, "%u) Empty ", i);
else {
sf0 = sf1 = sf2 = sf3 = sf4 = 0;
si0 = si1 = si2 = si3 = si4 = 0;
sw0 = sw1 = sw2 = sw3 = sw4 = 0;
for (j = 0; j < UMTX_CHAINS; j++) {
uc = &umtxq_chains[i][j];
mtx_lock(&uc->uc_lock);
whole = uc->max_length * 100;
mtx_unlock(&uc->uc_lock);
fract = (whole % tot) * 100;
if (UPROF_PERC_BIGGER(whole, fract, sw0, sf0)) {
sf0 = fract;
si0 = j;
sw0 = whole;
} else if (UPROF_PERC_BIGGER(whole, fract, sw1,
sf1)) {
sf1 = fract;
si1 = j;
sw1 = whole;
} else if (UPROF_PERC_BIGGER(whole, fract, sw2,
sf2)) {
sf2 = fract;
si2 = j;
sw2 = whole;
} else if (UPROF_PERC_BIGGER(whole, fract, sw3,
sf3)) {
sf3 = fract;
si3 = j;
sw3 = whole;
} else if (UPROF_PERC_BIGGER(whole, fract, sw4,
sf4)) {
sf4 = fract;
si4 = j;
sw4 = whole;
}
}
sbuf_printf(&sb, "queue %u:\n", i);
sbuf_printf(&sb, "1st: %u.%u%% idx: %u\n", sw0 / tot,
sf0 / tot, si0);
sbuf_printf(&sb, "2nd: %u.%u%% idx: %u\n", sw1 / tot,
sf1 / tot, si1);
sbuf_printf(&sb, "3rd: %u.%u%% idx: %u\n", sw2 / tot,
sf2 / tot, si2);
sbuf_printf(&sb, "4th: %u.%u%% idx: %u\n", sw3 / tot,
sf3 / tot, si3);
sbuf_printf(&sb, "5th: %u.%u%% idx: %u\n", sw4 / tot,
sf4 / tot, si4);
}
}
sbuf_trim(&sb);
sbuf_finish(&sb);
sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
sbuf_delete(&sb);
return (0);
}
static int
sysctl_debug_umtx_chains_clear(SYSCTL_HANDLER_ARGS)
{
struct umtxq_chain *uc;
u_int i, j;
int clear, error;
clear = 0;
error = sysctl_handle_int(oidp, &clear, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
if (clear != 0) {
for (i = 0; i < 2; ++i) {
for (j = 0; j < UMTX_CHAINS; ++j) {
uc = &umtxq_chains[i][j];
mtx_lock(&uc->uc_lock);
uc->length = 0;
uc->max_length = 0;
mtx_unlock(&uc->uc_lock);
}
}
}
return (0);
}
SYSCTL_PROC(_debug_umtx_chains, OID_AUTO, clear,
CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
sysctl_debug_umtx_chains_clear, "I", "Clear umtx chains statistics");
SYSCTL_PROC(_debug_umtx_chains, OID_AUTO, peaks,
CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0,
sysctl_debug_umtx_chains_peaks, "A", "Highest peaks in chains max length");
Add rudimentary profiling of the hash table used in the in the umtx code to hold active lock queues. Reviewed by: attilio Approved by: davidxu, gnn (mentor) MFC after: 3 weeks
2012-03-16 20:32:11 +00:00
#endif
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static void
umtxq_sysinit(void *arg __unused)
{
Use a seperated hash table for mutex and rwlock, avoid wasting some time on walking through idle threads sleeping on condition variables.
2008-05-30 02:18:54 +00:00
int i, j;
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx_pi_zone = uma_zcreate("umtx pi", sizeof(struct umtx_pi),
NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
Use a seperated hash table for mutex and rwlock, avoid wasting some time on walking through idle threads sleeping on condition variables.
2008-05-30 02:18:54 +00:00
for (i = 0; i < 2; ++i) {
for (j = 0; j < UMTX_CHAINS; ++j) {
mtx_init(&umtxq_chains[i][j].uc_lock, "umtxql", NULL,
MTX_DEF | MTX_DUPOK);
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
LIST_INIT(&umtxq_chains[i][j].uc_queue[0]);
LIST_INIT(&umtxq_chains[i][j].uc_queue[1]);
LIST_INIT(&umtxq_chains[i][j].uc_spare_queue);
Use a seperated hash table for mutex and rwlock, avoid wasting some time on walking through idle threads sleeping on condition variables.
2008-05-30 02:18:54 +00:00
TAILQ_INIT(&umtxq_chains[i][j].uc_pi_list);
umtxq_chains[i][j].uc_busy = 0;
umtxq_chains[i][j].uc_waiters = 0;
Fix some style bugs introduced in a previous commit (r233045) Reported by: glebius, jmallet Reviewed by: jmallet Approved by: gnn (mentor) MFC after: 2 days
2012-04-14 23:53:31 +00:00
#ifdef UMTX_PROFILING
Add rudimentary profiling of the hash table used in the in the umtx code to hold active lock queues. Reviewed by: attilio Approved by: davidxu, gnn (mentor) MFC after: 3 weeks
2012-03-16 20:32:11 +00:00
umtxq_chains[i][j].length = 0;
umtxq_chains[i][j].max_length = 0;
Fix some style bugs introduced in a previous commit (r233045) Reported by: glebius, jmallet Reviewed by: jmallet Approved by: gnn (mentor) MFC after: 2 days
2012-04-14 23:53:31 +00:00
#endif
Use a seperated hash table for mutex and rwlock, avoid wasting some time on walking through idle threads sleeping on condition variables.
2008-05-30 02:18:54 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
Fix some style bugs introduced in a previous commit (r233045) Reported by: glebius, jmallet Reviewed by: jmallet Approved by: gnn (mentor) MFC after: 2 days
2012-04-14 23:53:31 +00:00
#ifdef UMTX_PROFILING
Add rudimentary profiling of the hash table used in the in the umtx code to hold active lock queues. Reviewed by: attilio Approved by: davidxu, gnn (mentor) MFC after: 3 weeks
2012-03-16 20:32:11 +00:00
umtx_init_profiling();
Fix some style bugs introduced in a previous commit (r233045) Reported by: glebius, jmallet Reviewed by: jmallet Approved by: gnn (mentor) MFC after: 2 days
2012-04-14 23:53:31 +00:00
#endif
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_init(&umtx_lock, "umtx lock", NULL, MTX_DEF);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
EVENTHANDLER_REGISTER(process_exec, umtx_exec_hook, NULL,
EVENTHANDLER_PRI_ANY);
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
umtx_shm_init();
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
Allocate umtx_q from heap instead of stack, this avoids page fault panic in kernel under heavy swapping.
2005-03-05 09:15:03 +00:00
struct umtx_q *
umtxq_alloc(void)
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtx_q *uq;
uq = malloc(sizeof(struct umtx_q), M_UMTX, M_WAITOK | M_ZERO);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uq->uq_spare_queue = malloc(sizeof(struct umtxq_queue), M_UMTX,
M_WAITOK | M_ZERO);
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
TAILQ_INIT(&uq->uq_spare_queue->head);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
TAILQ_INIT(&uq->uq_pi_contested);
uq->uq_inherited_pri = PRI_MAX;
return (uq);
Allocate umtx_q from heap instead of stack, this avoids page fault panic in kernel under heavy swapping.
2005-03-05 09:15:03 +00:00
}
void
umtxq_free(struct umtx_q *uq)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
MPASS(uq->uq_spare_queue != NULL);
free(uq->uq_spare_queue, M_UMTX);
Allocate umtx_q from heap instead of stack, this avoids page fault panic in kernel under heavy swapping.
2005-03-05 09:15:03 +00:00
free(uq, M_UMTX);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static inline void
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
umtxq_hash(struct umtx_key *key)
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
unsigned n;
n = (uintptr_t)key->info.both.a + key->info.both.b;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
key->hash = ((n * GOLDEN_RATIO_PRIME) >> UMTX_SHIFTS) % UMTX_CHAINS;
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static inline struct umtxq_chain *
umtxq_getchain(struct umtx_key *key)
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
put semaphore waiter in long term list.
2010-01-09 06:12:44 +00:00
if (key->type <= TYPE_SEM)
Use a seperated hash table for mutex and rwlock, avoid wasting some time on walking through idle threads sleeping on condition variables.
2008-05-30 02:18:54 +00:00
return (&umtxq_chains[1][key->hash]);
return (&umtxq_chains[0][key->hash]);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
* Lock a chain.
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
*/
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
static inline void
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_lock(struct umtx_key *key)
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtxq_chain *uc;
uc = umtxq_getchain(key);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
mtx_lock(&uc->uc_lock);
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
* Unlock a chain.
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
*/
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
static inline void
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_unlock(struct umtx_key *key)
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtxq_chain *uc;
uc = umtxq_getchain(key);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
mtx_unlock(&uc->uc_lock);
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
* Set chain to busy state when following operation
* may be blocked (kernel mutex can not be used).
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
*/
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
static inline void
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_busy(struct umtx_key *key)
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtxq_chain *uc;
uc = umtxq_getchain(key);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
mtx_assert(&uc->uc_lock, MA_OWNED);
if (uc->uc_busy) {
let umtxq_busy() only spin on mp machine. make function name do_rwlock_unlock to be consistent with others.
2008-04-03 11:49:20 +00:00
#ifdef SMP
if (smp_cpus > 1) {
int count = BUSY_SPINS;
if (count > 0) {
umtxq_unlock(key);
while (uc->uc_busy && --count > 0)
cpu_spinwait();
umtxq_lock(key);
}
}
#endif
while (uc->uc_busy) {
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
uc->uc_waiters++;
msleep(uc, &uc->uc_lock, 0, "umtxqb", 0);
uc->uc_waiters--;
}
}
uc->uc_busy = 1;
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
* Unbusy a chain.
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
*/
Forgot to inline umtxq_unlock.
2004-11-30 12:18:53 +00:00
static inline void
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_unbusy(struct umtx_key *key)
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtxq_chain *uc;
uc = umtxq_getchain(key);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
mtx_assert(&uc->uc_lock, MA_OWNED);
KASSERT(uc->uc_busy != 0, ("not busy"));
uc->uc_busy = 0;
if (uc->uc_waiters)
wakeup_one(uc);
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
}
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
static inline void
umtxq_unbusy_unlocked(struct umtx_key *key)
{
umtxq_lock(key);
umtxq_unbusy(key);
umtxq_unlock(key);
}
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
static struct umtxq_queue *
umtxq_queue_lookup(struct umtx_key *key, int q)
{
struct umtxq_queue *uh;
struct umtxq_chain *uc;
uc = umtxq_getchain(key);
UMTXQ_LOCKED_ASSERT(uc);
LIST_FOREACH(uh, &uc->uc_queue[q], link) {
if (umtx_key_match(&uh->key, key))
return (uh);
}
return (NULL);
}
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
static inline void
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_insert_queue(struct umtx_q *uq, int q)
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
{
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
struct umtxq_queue *uh;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtxq_chain *uc;
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
uc = umtxq_getchain(&uq->uq_key);
UMTXQ_LOCKED_ASSERT(uc);
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
KASSERT((uq->uq_flags & UQF_UMTXQ) == 0, ("umtx_q is already on queue"));
In function umtxq_insert_queue, use parameter q (shared/exclusive queue) instead of hard coded constant. This does not affect RELENG_8 and previous, because the code only exists in the HEAD.
2010-02-10 05:47:34 +00:00
uh = umtxq_queue_lookup(&uq->uq_key, q);
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
if (uh != NULL) {
LIST_INSERT_HEAD(&uc->uc_spare_queue, uq->uq_spare_queue, link);
} else {
uh = uq->uq_spare_queue;
uh->key = uq->uq_key;
LIST_INSERT_HEAD(&uc->uc_queue[q], uh, link);
Fix a bug in UMTX_PROFILING: UMTX_PROFILING should really analyze the distribution of locks as they index entries in the umtxq_chains hash-table. However, the current implementation does add/dec the length counters for *every* thread insert/removal, measuring at all really userland contention and not the hash distribution. Fix this by correctly add/dec the length counters in the points where it is really needed. Please note that this bug brought us questioning in the past the quality of the umtx hash table distribution. To date with all the benchmarks I could try I was not able to reproduce any issue about the hash distribution on umtx. Sponsored by: EMC / Isilon storage division Reviewed by: jeff, davide MFC after: 2 weeks
2013-03-21 19:58:25 +00:00
#ifdef UMTX_PROFILING
uc->length++;
if (uc->length > uc->max_length) {
uc->max_length = uc->length;
if (uc->max_length > max_length)
max_length = uc->max_length;
}
#endif
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
}
uq->uq_spare_queue = NULL;
TAILQ_INSERT_TAIL(&uh->head, uq, uq_link);
uh->length++;
Move flag TDF_UMTXQ into structure umtxq, this eliminates the requirement of scheduler lock in some umtx code.
2006-05-18 08:43:46 +00:00
uq->uq_flags |= UQF_UMTXQ;
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
uq->uq_cur_queue = uh;
return;
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
static inline void
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_remove_queue(struct umtx_q *uq, int q)
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtxq_chain *uc;
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
struct umtxq_queue *uh;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
uc = umtxq_getchain(&uq->uq_key);
UMTXQ_LOCKED_ASSERT(uc);
Move flag TDF_UMTXQ into structure umtxq, this eliminates the requirement of scheduler lock in some umtx code.
2006-05-18 08:43:46 +00:00
if (uq->uq_flags & UQF_UMTXQ) {
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
uh = uq->uq_cur_queue;
TAILQ_REMOVE(&uh->head, uq, uq_link);
uh->length--;
Move flag TDF_UMTXQ into structure umtxq, this eliminates the requirement of scheduler lock in some umtx code.
2006-05-18 08:43:46 +00:00
uq->uq_flags &= ~UQF_UMTXQ;
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
if (TAILQ_EMPTY(&uh->head)) {
KASSERT(uh->length == 0,
("inconsistent umtxq_queue length"));
Fix a bug in UMTX_PROFILING: UMTX_PROFILING should really analyze the distribution of locks as they index entries in the umtxq_chains hash-table. However, the current implementation does add/dec the length counters for *every* thread insert/removal, measuring at all really userland contention and not the hash distribution. Fix this by correctly add/dec the length counters in the points where it is really needed. Please note that this bug brought us questioning in the past the quality of the umtx hash table distribution. To date with all the benchmarks I could try I was not able to reproduce any issue about the hash distribution on umtx. Sponsored by: EMC / Isilon storage division Reviewed by: jeff, davide MFC after: 2 weeks
2013-03-21 19:58:25 +00:00
#ifdef UMTX_PROFILING
uc->length--;
#endif
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
LIST_REMOVE(uh, link);
} else {
uh = LIST_FIRST(&uc->uc_spare_queue);
KASSERT(uh != NULL, ("uc_spare_queue is empty"));
LIST_REMOVE(uh, link);
}
uq->uq_spare_queue = uh;
uq->uq_cur_queue = NULL;
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Check if there are multiple waiters
*/
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
static int
umtxq_count(struct umtx_key *key)
{
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
struct umtxq_queue *uh;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx: don't call umtxq_getchain unless the value is needed
2018-05-19 05:09:10 +00:00
UMTXQ_LOCKED_ASSERT(umtxq_getchain(key));
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
uh = umtxq_queue_lookup(key, UMTX_SHARED_QUEUE);
if (uh != NULL)
return (uh->length);
return (0);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
/*
* Check if there are multiple PI waiters and returns first
* waiter.
*/
static int
umtxq_count_pi(struct umtx_key *key, struct umtx_q **first)
{
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
struct umtxq_queue *uh;
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
*first = NULL;
umtx: don't call umtxq_getchain unless the value is needed
2018-05-19 05:09:10 +00:00
UMTXQ_LOCKED_ASSERT(umtxq_getchain(key));
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
uh = umtxq_queue_lookup(key, UMTX_SHARED_QUEUE);
if (uh != NULL) {
*first = TAILQ_FIRST(&uh->head);
return (uh->length);
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
}
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
return (0);
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
/*
* Check for possible stops and suspensions while executing a umtx
* locking operation.
*
* The sleep argument controls whether the function can handle a stop
* request itself or it should return ERESTART and the request is
* proceed at the kernel/user boundary in ast.
*
* Typically, when retrying due to casueword(9) failure (rv == 1), we
* should handle the stop requests there, with exception of cases when
* the thread busied the umtx key, or when functions return
* immediately if umtxq_check_susp() returned non-zero. On the other
* hand, retrying the whole lock operation, we better not stop there
* but delegate the handling to ast.
*
* If the request is for thread termination P_SINGLE_EXIT, we cannot
* handle it at all, and simply return EINTR.
*/
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
static int
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
umtxq_check_susp(struct thread *td, bool sleep)
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
{
struct proc *p;
int error;
/*
* The check for TDF_NEEDSUSPCHK is racy, but it is enough to
* eventually break the lockstep loop.
*/
if ((td->td_flags & TDF_NEEDSUSPCHK) == 0)
return (0);
error = 0;
p = td->td_proc;
PROC_LOCK(p);
Make umtxq_check_susp() to correctly handle thread exit requests. The check for P_SINGLE_EXIT was shadowed by the (P_SHOULDSTOP || traced) check. Reported by: bdrewery (might be) Reviewed by: markj Tested by: pho MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential revision: https://reviews.freebsd.org/D21124
2019-08-01 14:34:27 +00:00
if (p->p_flag & P_SINGLE_EXIT)
error = EINTR;
else if (P_SHOULDSTOP(p) ||
((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND)))
error = sleep ? thread_suspend_check(0) : ERESTART;
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
PROC_UNLOCK(p);
return (error);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Wake up threads waiting on an userland object.
*/
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
static int
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_signal_queue(struct umtx_key *key, int n_wake, int q)
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
{
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
struct umtxq_queue *uh;
struct umtx_q *uq;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
int ret;
- Remove the blocked pointer from the umtx structure. - Use a hash of umtx queues to queue blocked threads. We hash on pid and the virtual address of the umtx structure. This eliminates cases where we previously held a lock across a casuptr call. Reviwed by: jhb (quickly)
2003-06-03 05:24:46 +00:00
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
ret = 0;
umtx: don't call umtxq_getchain unless the value is needed
2018-05-19 05:09:10 +00:00
UMTXQ_LOCKED_ASSERT(umtxq_getchain(key));
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
uh = umtxq_queue_lookup(key, q);
if (uh != NULL) {
while ((uq = TAILQ_FIRST(&uh->head)) != NULL) {
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_remove_queue(uq, q);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
wakeup(uq);
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
if (++ret >= n_wake)
Make a chain be a list of queues, and make threads waiting for same key coalesce to same queue, this makes searching path shorter and improves performance. Also fix comments about shared PI-mutex.
2010-01-10 09:31:57 +00:00
return (ret);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
}
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
return (ret);
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Wake up specified thread.
*/
static inline void
umtxq_signal_thread(struct umtx_q *uq)
{
umtx: don't call umtxq_getchain unless the value is needed
2018-05-19 05:09:10 +00:00
UMTXQ_LOCKED_ASSERT(umtxq_getchain(&uq->uq_key));
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_remove(uq);
wakeup(uq);
}
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
static inline int
tstohz(const struct timespec *tsp)
{
struct timeval tv;
TIMESPEC_TO_TIMEVAL(&tv, tsp);
return tvtohz(&tv);
}
static void
abs_timeout_init(struct abs_timeout *timo, int clockid, int absolute,
const struct timespec *timeout)
{
timo->clockid = clockid;
if (!absolute) {
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
timo->is_abs_real = false;
abs_timeout_update(timo);
Make timespecadd(3) and friends public The timespecadd(3) family of macros were imported from NetBSD back in r35029. However, they were initially guarded by #ifdef _KERNEL. In the meantime, we have grown at least 28 syscalls that use timespecs in some way, leading many programs both inside and outside of the base system to redefine those macros. It's better just to make the definitions public. Our kernel currently defines two-argument versions of timespecadd and timespecsub. NetBSD, OpenBSD, and FreeDesktop.org's libbsd, however, define three-argument versions. Solaris also defines a three-argument version, but only in its kernel. This revision changes our definition to match the common three-argument version. Bump _FreeBSD_version due to the breaking KPI change. Discussed with: cem, jilles, ian, bde Differential Revision: https://reviews.freebsd.org/D14725
2018-07-30 15:46:40 +00:00
timespecadd(&timo->cur, timeout, &timo->end);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
} else {
timo->end = *timeout;
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
timo->is_abs_real = clockid == CLOCK_REALTIME ||
clockid == CLOCK_REALTIME_FAST ||
clockid == CLOCK_REALTIME_PRECISE;
/*
* If is_abs_real, umtxq_sleep will read the clock
* after setting td_rtcgen; otherwise, read it here.
*/
if (!timo->is_abs_real) {
abs_timeout_update(timo);
}
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
}
}
static void
abs_timeout_init2(struct abs_timeout *timo, const struct _umtx_time *umtxtime)
{
abs_timeout_init(timo, umtxtime->_clockid,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
(umtxtime->_flags & UMTX_ABSTIME) != 0, &umtxtime->_timeout);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
}
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
static inline void
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
abs_timeout_update(struct abs_timeout *timo)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
kern_clock_gettime(curthread, timo->clockid, &timo->cur);
}
static int
abs_timeout_gethz(struct abs_timeout *timo)
{
struct timespec tts;
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
if (timespeccmp(&timo->end, &timo->cur, <=))
return (-1);
Make timespecadd(3) and friends public The timespecadd(3) family of macros were imported from NetBSD back in r35029. However, they were initially guarded by #ifdef _KERNEL. In the meantime, we have grown at least 28 syscalls that use timespecs in some way, leading many programs both inside and outside of the base system to redefine those macros. It's better just to make the definitions public. Our kernel currently defines two-argument versions of timespecadd and timespecsub. NetBSD, OpenBSD, and FreeDesktop.org's libbsd, however, define three-argument versions. Solaris also defines a three-argument version, but only in its kernel. This revision changes our definition to match the common three-argument version. Bump _FreeBSD_version due to the breaking KPI change. Discussed with: cem, jilles, ian, bde Differential Revision: https://reviews.freebsd.org/D14725
2018-07-30 15:46:40 +00:00
timespecsub(&timo->end, &timo->cur, &tts);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
return (tstohz(&tts));
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
static uint32_t
umtx_unlock_val(uint32_t flags, bool rb)
{
if (rb)
return (UMUTEX_RB_OWNERDEAD);
else if ((flags & UMUTEX_NONCONSISTENT) != 0)
return (UMUTEX_RB_NOTRECOV);
else
return (UMUTEX_UNOWNED);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Put thread into sleep state, before sleeping, check if
* thread was removed from umtx queue.
*/
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
static inline int
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
umtxq_sleep(struct umtx_q *uq, const char *wmesg, struct abs_timeout *abstime)
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtxq_chain *uc;
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
int error, timo;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
if (abstime != NULL && abstime->is_abs_real) {
curthread->td_rtcgen = atomic_load_acq_int(&rtc_generation);
abs_timeout_update(abstime);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
uc = umtxq_getchain(&uq->uq_key);
UMTXQ_LOCKED_ASSERT(uc);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
for (;;) {
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
if (!(uq->uq_flags & UQF_UMTXQ)) {
error = 0;
break;
}
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
if (abstime != NULL) {
timo = abs_timeout_gethz(abstime);
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
if (timo < 0) {
error = ETIMEDOUT;
break;
}
tvtohz will print out an error message if a negative value is given to it, avoid this problem by detecting timeout earlier. Reported by: pho
2012-08-11 00:06:56 +00:00
} else
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
timo = 0;
error = msleep(uq, &uc->uc_lock, PCATCH | PDROP, wmesg, timo);
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
if (error == EINTR || error == ERESTART) {
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
umtxq_lock(&uq->uq_key);
break;
}
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
if (abstime != NULL) {
if (abstime->is_abs_real)
curthread->td_rtcgen =
atomic_load_acq_int(&rtc_generation);
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
abs_timeout_update(abstime);
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
}
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
umtxq_lock(&uq->uq_key);
}
When the RTC is adjusted, reevaluate absolute sleep times based on the RTC POSIX 2008 says this about clock_settime(2): If the value of the CLOCK_REALTIME clock is set via clock_settime(), the new value of the clock shall be used to determine the time of expiration for absolute time services based upon the CLOCK_REALTIME clock. This applies to the time at which armed absolute timers expire. If the absolute time requested at the invocation of such a time service is before the new value of the clock, the time service shall expire immediately as if the clock had reached the requested time normally. Setting the value of the CLOCK_REALTIME clock via clock_settime() shall have no effect on threads that are blocked waiting for a relative time service based upon this clock, including the nanosleep() function; nor on the expiration of relative timers based upon this clock. Consequently, these time services shall expire when the requested relative interval elapses, independently of the new or old value of the clock. When the real-time clock is adjusted, such as by clock_settime(3), wake any threads sleeping until an absolute real-clock time. Such a sleep is indicated by a non-zero td_rtcgen. The sleep functions will set that field to zero and return zero to tell the caller to reevaluate its sleep duration based on the new value of the clock. At present, this affects the following functions: pthread_cond_timedwait(3) pthread_mutex_timedlock(3) pthread_rwlock_timedrdlock(3) pthread_rwlock_timedwrlock(3) sem_timedwait(3) sem_clockwait_np(3) I'm working on adding clock_nanosleep(2), which will also be affected. Reported by: Sebastian Huber <sebastian.huber@embedded-brains.de> Reviewed by: jhb, kib MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9791
2017-03-14 19:06:44 +00:00
curthread->td_rtcgen = 0;
Return ETIMEDOUT when thread is timeouted since POSIX thread APIs expect ETIMEDOUT not EAGAIN, this simplifies userland code a bit.
2005-01-06 02:08:34 +00:00
return (error);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Convert userspace address into unique logical address.
*/
Expose the umtx_key structure and API to the rest of the kernel. MFC after: 3 days
2011-02-23 13:19:14 +00:00
int
Add missing const keyword to function parameter. The umtx_key_get() function does not dereference the address off the userspace object. The pointer can safely be const.
2015-08-03 21:11:33 +00:00
umtx_key_get(const void *addr, int type, int share, struct umtx_key *key)
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct thread *td = curthread;
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
vm_map_t map;
vm_map_entry_t entry;
vm_pindex_t pindex;
vm_prot_t prot;
boolean_t wired;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
key->type = type;
if (share == THREAD_SHARE) {
key->shared = 0;
key->info.private.vs = td->td_proc->p_vmspace;
key->info.private.addr = (uintptr_t)addr;
Eliminate an unnecessary `if' statement.
2006-10-25 06:28:23 +00:00
} else {
MPASS(share == PROCESS_SHARE || share == AUTO_SHARE);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
map = &td->td_proc->p_vmspace->vm_map;
if (vm_map_lookup(&map, (vm_offset_t)addr, VM_PROT_WRITE,
&entry, &key->info.shared.object, &pindex, &prot,
&wired) != KERN_SUCCESS) {
Minor (and incomplete) style cleanup. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 20:47:42 +00:00
return (EFAULT);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
Axe unused code.
2006-02-04 06:36:39 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if ((share == PROCESS_SHARE) ||
(share == AUTO_SHARE &&
VM_INHERIT_SHARE == entry->inheritance)) {
key->shared = 1;
Fix bad arithmetic in umtx_key_get() to compute object offset. It looks like umtx_key_get() has the addition and subtraction the wrong way around, meaning that it fails to match in certain cases. This causes the cloudlibc unit tests to deadlock in certain cases. Reviewed by: kib Differential Revision: https://reviews.freebsd.org/D3287
2015-08-04 06:01:13 +00:00
key->info.shared.offset = (vm_offset_t)addr -
entry->start + entry->offset;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
vm_object_reference(key->info.shared.object);
} else {
key->shared = 0;
key->info.private.vs = td->td_proc->p_vmspace;
key->info.private.addr = (uintptr_t)addr;
}
vm_map_lookup_done(map, entry);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_hash(key);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
return (0);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Release key.
*/
Expose the umtx_key structure and API to the rest of the kernel. MFC after: 3 days
2011-02-23 13:19:14 +00:00
void
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
umtx_key_release(struct umtx_key *key)
{
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (key->shared)
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
vm_object_deallocate(key->info.shared.object);
1. use per-chain mutex instead of global mutex to reduce lock collision. 2. Fix two race conditions. One is between _umtx_unlock and signal, also a thread was marked TDF_UMTXWAKEUP by _umtx_unlock, it is possible a signal delivered to the thread will cause msleep returns EINTR, and the thread breaks out of loop, this causes umtx ownership is not transfered to the thread. Another is in _umtx_unlock itself, when the function sets the umtx to UMTX_UNOWNED state, a new thread can come in and lock the umtx, also the function tries to set contested bit flag, but it will fail. Although the function will wake a blocked thread, if that thread breaks out of loop by signal, no contested bit will be set.
2004-11-30 12:02:53 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Fetch and compare value, sleep on the address if value is not changed.
*/
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
static int
o Add keyword volatile for user mutex owner field. o Fix type consistent problem by using type long for old umtx and wait channel. o Rename casuptr to casuword.
2006-10-17 02:24:47 +00:00
do_wait(struct thread *td, void *addr, u_long id,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
struct _umtx_time *timeout, int compat32, int is_private)
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
{
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout timo;
Allocate umtx_q from heap instead of stack, this avoids page fault panic in kernel under heavy swapping.
2005-03-05 09:15:03 +00:00
struct umtx_q *uq;
o Add keyword volatile for user mutex owner field. o Fix type consistent problem by using type long for old umtx and wait channel. o Rename casuptr to casuword.
2006-10-17 02:24:47 +00:00
u_long tmp;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
uint32_t tmp32;
make umtx timeout relative so userland can select different clock type, e.g, CLOCK_REALTIME or CLOCK_MONOTONIC. merge umtx_wait and umtx_timedwait into single function.
2005-01-14 13:38:15 +00:00
int error = 0;
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
Allocate umtx_q from heap instead of stack, this avoids page fault panic in kernel under heavy swapping.
2005-03-05 09:15:03 +00:00
uq = td->td_umtxq;
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
if ((error = umtx_key_get(addr, TYPE_SIMPLE_WAIT,
is_private ? THREAD_SHARE : AUTO_SHARE, &uq->uq_key)) != 0)
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
return (error);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
if (timeout != NULL)
abs_timeout_init2(&timo, timeout);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_lock(&uq->uq_key);
umtxq_insert(uq);
umtxq_unlock(&uq->uq_key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (compat32 == 0) {
error = fueword(addr, &tmp);
if (error != 0)
error = EFAULT;
} else {
error = fueword32(addr, &tmp32);
if (error == 0)
tmp = tmp32;
else
error = EFAULT;
}
Reduce code size by creating common timed sleeping function.
2012-03-29 02:46:43 +00:00
umtxq_lock(&uq->uq_key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == 0) {
if (tmp == id)
error = umtxq_sleep(uq, "uwait", timeout == NULL ?
NULL : &timo);
if ((uq->uq_flags & UQF_UMTXQ) == 0)
error = 0;
else
umtxq_remove(uq);
} else if ((uq->uq_flags & UQF_UMTXQ) != 0) {
Reduce code size by creating common timed sleeping function.
2012-03-29 02:46:43 +00:00
umtxq_remove(uq);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
}
Reduce code size by creating common timed sleeping function.
2012-03-29 02:46:43 +00:00
umtxq_unlock(&uq->uq_key);
Allocate umtx_q from heap instead of stack, this avoids page fault panic in kernel under heavy swapping.
2005-03-05 09:15:03 +00:00
umtx_key_release(&uq->uq_key);
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
if (error == ERESTART)
error = EINTR;
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
return (error);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Wake up threads sleeping on the specified address.
*/
do umtx_wake at userland thread exit address, so that others userland threads can wait for a thread to exit, and safely assume that the thread has left userland and is no longer using its userland stack, this is necessary for pthread_join when a thread is waiting for another thread to exit which has user customized stack, after pthread_join returns, the userland stack can be reused for other purposes, without this change, the joiner thread has to spin at the address to ensure the thread is really exited.
2005-10-26 06:55:46 +00:00
int
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
kern_umtx_wake(struct thread *td, void *uaddr, int n_wake, int is_private)
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
{
struct umtx_key key;
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
int ret;
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
if ((ret = umtx_key_get(uaddr, TYPE_SIMPLE_WAIT,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
is_private ? THREAD_SHARE : AUTO_SHARE, &key)) != 0)
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
return (ret);
Add umtxq_lock/unlock around umtx_signal, fix debug kernel compiling, let umtx_lock returns EINTR when it returns ERESTART, this lets userland have chance to back off mtx lock code when needed.
2004-12-24 11:59:20 +00:00
umtxq_lock(&key);
Clean up some cosmetic nits in kern_umtx.c, found during recent work in this area and by the Clang static analyzer. Remove some dead assignments. Fix a typo in a panic string. Use umtx_pi_disown() instead of duplicate code. Use an existing variable instead of curthread. Approved by: kib (mentor) MFC after: 3 days Sponsored by: Dell Inc
2015-03-28 21:21:40 +00:00
umtxq_signal(&key, n_wake);
Add umtxq_lock/unlock around umtx_signal, fix debug kernel compiling, let umtx_lock returns EINTR when it returns ERESTART, this lets userland have chance to back off mtx lock code when needed.
2004-12-24 11:59:20 +00:00
umtxq_unlock(&key);
1. Fix race condition between umtx lock and unlock, heavy testing on SMP can explore the bug. 2. Let umtx_wake returns number of threads have been woken.
2004-12-24 11:30:55 +00:00
umtx_key_release(&key);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
return (0);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Lock PTHREAD_PRIO_NONE protocol POSIX mutex.
*/
static int
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
do_lock_normal(struct thread *td, struct umutex *m, uint32_t flags,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
struct _umtx_time *timeout, int mode)
Add user priority loaning code to support priority propagation for 1:1 threading's POSIX priority mutexes, the code is no-op unless priority-aware umtx code is committed.
2006-08-25 06:12:53 +00:00
{
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout timo;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtx_q *uq;
uint32_t owner, old, id;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
int error, rv;
Add user priority loaning code to support priority propagation for 1:1 threading's POSIX priority mutexes, the code is no-op unless priority-aware umtx code is committed.
2006-08-25 06:12:53 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
id = td->td_tid;
uq = td->td_umtxq;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = 0;
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
if (timeout != NULL)
abs_timeout_init2(&timo, timeout);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Care must be exercised when dealing with umtx structure. It
* can fault on any access.
*/
for (;;) {
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&m->m_owner, &owner);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1)
return (EFAULT);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
if (mode == _UMUTEX_WAIT) {
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (owner == UMUTEX_UNOWNED ||
owner == UMUTEX_CONTESTED ||
owner == UMUTEX_RB_OWNERDEAD ||
owner == UMUTEX_RB_NOTRECOV)
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
return (0);
} else {
/*
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
* Robust mutex terminated. Kernel duty is to
* return EOWNERDEAD to the userspace. The
* umutex.m_flags UMUTEX_NONCONSISTENT is set
* by the common userspace code.
*/
if (owner == UMUTEX_RB_OWNERDEAD) {
rv = casueword32(&m->m_owner,
UMUTEX_RB_OWNERDEAD, &owner,
id | UMUTEX_CONTESTED);
if (rv == -1)
return (EFAULT);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(owner == UMUTEX_RB_OWNERDEAD);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (EOWNERDEAD); /* success */
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
MPASS(rv == 1);
rv = umtxq_check_susp(td, false);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (rv != 0)
return (rv);
continue;
}
if (owner == UMUTEX_RB_NOTRECOV)
return (ENOTRECOVERABLE);
/*
* Try the uncontested case. This should be
* done in userland.
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
*/
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&m->m_owner, UMUTEX_UNOWNED,
&owner, id);
/* The address was invalid. */
if (rv == -1)
return (EFAULT);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
/* The acquire succeeded. */
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(owner == UMUTEX_UNOWNED);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (0);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
/*
* If no one owns it but it is contested try
* to acquire it.
*/
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
MPASS(rv == 1);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
if (owner == UMUTEX_CONTESTED) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&m->m_owner,
UMUTEX_CONTESTED, &owner,
id | UMUTEX_CONTESTED);
/* The address was invalid. */
if (rv == -1)
return (EFAULT);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(owner == UMUTEX_CONTESTED);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
return (0);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
if (rv == 1) {
rv = umtxq_check_susp(td, false);
if (rv != 0)
return (rv);
}
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
/*
* If this failed the lock has
* changed, restart.
*/
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
continue;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
/* rv == 1 but not contested, likely store failure */
rv = umtxq_check_susp(td, false);
if (rv != 0)
return (rv);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
if (mode == _UMUTEX_TRY)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EBUSY);
/*
* If we caught a signal, we have retried and now
* exit immediately.
*/
Remove trailing space.
2012-03-30 05:49:32 +00:00
if (error != 0)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (error);
if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX,
GET_SHARE(flags), &uq->uq_key)) != 0)
return (error);
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_insert(uq);
umtxq_unlock(&uq->uq_key);
/*
* Set the contested bit so that a release in user space
* knows to use the system call for unlock. If this fails
* either some one else has acquired the lock or it has been
* released.
*/
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&m->m_owner, owner, &old,
owner | UMUTEX_CONTESTED);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
/* The address was invalid or casueword failed to store. */
if (rv == -1 || rv == 1) {
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_lock(&uq->uq_key);
umtxq_remove(uq);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
umtxq_unbusy(&uq->uq_key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == -1)
return (EFAULT);
if (rv == 1) {
rv = umtxq_check_susp(td, false);
if (rv != 0)
return (rv);
}
continue;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
/*
* We set the contested bit, sleep. Otherwise the lock changed
* and we need to retry or we lost a race to the thread
* unlocking the umtx.
*/
umtxq_lock(&uq->uq_key);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
umtxq_unbusy(&uq->uq_key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
MPASS(old == owner);
error = umtxq_sleep(uq, "umtxn", timeout == NULL ?
NULL : &timo);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error == 0)
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, false);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
return (0);
}
/*
* Unlock PTHREAD_PRIO_NONE protocol POSIX mutex.
*/
static int
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
do_unlock_normal(struct thread *td, struct umutex *m, uint32_t flags, bool rb)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
struct umtx_key key;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uint32_t owner, old, id, newlock;
int error, count;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
id = td->td_tid;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
again:
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Make sure we own this mtx.
*/
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = fueword32(&m->m_owner, &owner);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
if ((owner & ~UMUTEX_CONTESTED) != id)
return (EPERM);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
newlock = umtx_unlock_val(flags, rb);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if ((owner & UMUTEX_CONTESTED) == 0) {
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
error = casueword32(&m->m_owner, owner, &old, newlock);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (error == 1) {
error = umtxq_check_susp(td, false);
if (error != 0)
return (error);
goto again;
}
MPASS(old == owner);
return (0);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
/* We should only ever be in here for contested locks */
if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags),
&key)) != 0)
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
count = umtxq_count(&key);
umtxq_unlock(&key);
/*
* When unlocking the umtx, it must be marked as unowned if
* there is zero or one thread only waiting for it.
* Otherwise, it must be marked as contested.
*/
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (count > 1)
newlock |= UMUTEX_CONTESTED;
error = casueword32(&m->m_owner, owner, &old, newlock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_lock(&key);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
umtxq_signal(&key, 1);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (error == 1) {
if (old != owner)
return (EINVAL);
error = umtxq_check_susp(td, false);
if (error != 0)
return (error);
goto again;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (0);
}
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
/*
* Check if the mutex is available and wake up a waiter,
* only for simple mutex.
*/
static int
do_wake_umutex(struct thread *td, struct umutex *m)
{
struct umtx_key key;
uint32_t owner;
uint32_t flags;
int error;
int count;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
again:
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = fueword32(&m->m_owner, &owner);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
return (EFAULT);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if ((owner & ~UMUTEX_CONTESTED) != 0 && owner != UMUTEX_RB_OWNERDEAD &&
owner != UMUTEX_RB_NOTRECOV)
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
return (0);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&m->m_flags, &flags);
if (error == -1)
return (EFAULT);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
/* We should only ever be in here for contested locks */
if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags),
&key)) != 0)
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
count = umtxq_count(&key);
umtxq_unlock(&key);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (count <= 1 && owner != UMUTEX_RB_OWNERDEAD &&
owner != UMUTEX_RB_NOTRECOV) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = casueword32(&m->m_owner, UMUTEX_CONTESTED, &owner,
UMUTEX_UNOWNED);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (error == -1) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = EFAULT;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
} else if (error == 1) {
umtxq_lock(&key);
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
error = umtxq_check_susp(td, false);
if (error != 0)
return (error);
goto again;
}
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
}
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
umtxq_lock(&key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (error == 0 && count != 0) {
MPASS((owner & ~UMUTEX_CONTESTED) == 0 ||
owner == UMUTEX_RB_OWNERDEAD ||
owner == UMUTEX_RB_NOTRECOV);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
umtxq_signal(&key, 1);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
return (error);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
}
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
/*
* Check if the mutex has waiters and tries to fix contention bit.
*/
static int
do_wake2_umutex(struct thread *td, struct umutex *m, uint32_t flags)
{
struct umtx_key key;
uint32_t owner, old;
int type;
int error;
int count;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
switch (flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT |
UMUTEX_ROBUST)) {
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
case 0:
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
case UMUTEX_ROBUST:
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
type = TYPE_NORMAL_UMUTEX;
break;
case UMUTEX_PRIO_INHERIT:
type = TYPE_PI_UMUTEX;
break;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
case (UMUTEX_PRIO_INHERIT | UMUTEX_ROBUST):
type = TYPE_PI_ROBUST_UMUTEX;
break;
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
case UMUTEX_PRIO_PROTECT:
type = TYPE_PP_UMUTEX;
break;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
case (UMUTEX_PRIO_PROTECT | UMUTEX_ROBUST):
type = TYPE_PP_ROBUST_UMUTEX;
break;
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
default:
return (EINVAL);
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if ((error = umtx_key_get(m, type, GET_SHARE(flags), &key)) != 0)
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
return (error);
owner = 0;
umtxq_lock(&key);
umtxq_busy(&key);
count = umtxq_count(&key);
umtxq_unlock(&key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = fueword32(&m->m_owner, &owner);
if (error == -1)
error = EFAULT;
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
/*
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
* Only repair contention bit if there is a waiter, this means
* the mutex is still being referenced by userland code,
* otherwise don't update any memory.
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
*/
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
while (error == 0 && (owner & UMUTEX_CONTESTED) == 0 &&
(count > 1 || (count == 1 && (owner & ~UMUTEX_CONTESTED) != 0))) {
error = casueword32(&m->m_owner, owner, &old,
owner | UMUTEX_CONTESTED);
if (error == -1) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = EFAULT;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
break;
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (error == 0) {
MPASS(old == owner);
break;
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
owner = old;
error = umtxq_check_susp(td, false);
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
umtxq_lock(&key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == EFAULT) {
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
umtxq_signal(&key, INT_MAX);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
} else if (count != 0 && ((owner & ~UMUTEX_CONTESTED) == 0 ||
owner == UMUTEX_RB_OWNERDEAD || owner == UMUTEX_RB_NOTRECOV))
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
umtxq_signal(&key, 1);
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
return (error);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
static inline struct umtx_pi *
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
umtx_pi_alloc(int flags)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
struct umtx_pi *pi;
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
pi = uma_zalloc(umtx_pi_zone, M_ZERO | flags);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
TAILQ_INIT(&pi->pi_blocked);
atomic_add_int(&umtx_pi_allocated, 1);
return (pi);
}
static inline void
umtx_pi_free(struct umtx_pi *pi)
{
uma_zfree(umtx_pi_zone, pi);
atomic_add_int(&umtx_pi_allocated, -1);
}
/*
* Adjust the thread's position on a pi_state after its priority has been
* changed.
*/
static int
umtx_pi_adjust_thread(struct umtx_pi *pi, struct thread *td)
{
struct umtx_q *uq, *uq1, *uq2;
struct thread *td1;
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
mtx_assert(&umtx_lock, MA_OWNED);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (pi == NULL)
return (0);
uq = td->td_umtxq;
/*
* Check if the thread needs to be moved on the blocked chain.
* It needs to be moved if either its priority is lower than
* the previous thread or higher than the next thread.
*/
uq1 = TAILQ_PREV(uq, umtxq_head, uq_lockq);
uq2 = TAILQ_NEXT(uq, uq_lockq);
if ((uq1 != NULL && UPRI(td) < UPRI(uq1->uq_thread)) ||
(uq2 != NULL && UPRI(td) > UPRI(uq2->uq_thread))) {
/*
* Remove thread from blocked chain and determine where
* it should be moved to.
*/
TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq);
TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) {
td1 = uq1->uq_thread;
MPASS(td1->td_proc->p_magic == P_MAGIC);
if (UPRI(td1) > UPRI(td))
break;
}
if (uq1 == NULL)
TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq);
else
TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq);
}
return (1);
}
The dependency chain for priority-inheritance mutexes could be subverted by userspace into cycle. Both umtx_propagate_priority() and umtx_repropagate_priority() would then loop infinitely, owning the spinlock. Check for the cycle using standard Floyd' algorithm before doing the pass in the affected functions. Add simple check for condition of tricking the thread into a wait for itself, which could be easily simulated by usermode without race. Found by: Eric van Gyzen <eric@vangyzen.net> In collaboration with: Eric van Gyzen <eric@vangyzen.net> Tested by: pho MFC after: 1 week
2015-01-31 12:27:40 +00:00
static struct umtx_pi *
umtx_pi_next(struct umtx_pi *pi)
{
struct umtx_q *uq_owner;
if (pi->pi_owner == NULL)
return (NULL);
uq_owner = pi->pi_owner->td_umtxq;
if (uq_owner == NULL)
return (NULL);
return (uq_owner->uq_pi_blocked);
}
/*
* Floyd's Cycle-Finding Algorithm.
*/
static bool
umtx_pi_check_loop(struct umtx_pi *pi)
{
struct umtx_pi *pi1; /* fast iterator */
mtx_assert(&umtx_lock, MA_OWNED);
if (pi == NULL)
return (false);
pi1 = pi;
for (;;) {
pi = umtx_pi_next(pi);
if (pi == NULL)
break;
pi1 = umtx_pi_next(pi1);
if (pi1 == NULL)
break;
pi1 = umtx_pi_next(pi1);
if (pi1 == NULL)
break;
if (pi == pi1)
return (true);
}
return (false);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Propagate priority when a thread is blocked on POSIX
* PI mutex.
*/
static void
umtx_propagate_priority(struct thread *td)
{
struct umtx_q *uq;
struct umtx_pi *pi;
int pri;
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
mtx_assert(&umtx_lock, MA_OWNED);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
pri = UPRI(td);
uq = td->td_umtxq;
pi = uq->uq_pi_blocked;
if (pi == NULL)
return;
The dependency chain for priority-inheritance mutexes could be subverted by userspace into cycle. Both umtx_propagate_priority() and umtx_repropagate_priority() would then loop infinitely, owning the spinlock. Check for the cycle using standard Floyd' algorithm before doing the pass in the affected functions. Add simple check for condition of tricking the thread into a wait for itself, which could be easily simulated by usermode without race. Found by: Eric van Gyzen <eric@vangyzen.net> In collaboration with: Eric van Gyzen <eric@vangyzen.net> Tested by: pho MFC after: 1 week
2015-01-31 12:27:40 +00:00
if (umtx_pi_check_loop(pi))
return;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
for (;;) {
td = pi->pi_owner;
MFp4: It is possible a lower priority thread lending priority to higher priority thread, in old code, it is ignored, however the lending should always be recorded, add field td_lend_user_pri to fix the problem, if a thread does not have borrowed priority, its value is PRI_MAX. MFC after: 1 week
2010-12-09 02:42:02 +00:00
if (td == NULL || td == curthread)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return;
MPASS(td->td_proc != NULL);
MPASS(td->td_proc->p_magic == P_MAGIC);
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_lock(td);
MFp4: It is possible a lower priority thread lending priority to higher priority thread, in old code, it is ignored, however the lending should always be recorded, add field td_lend_user_pri to fix the problem, if a thread does not have borrowed priority, its value is PRI_MAX. MFC after: 1 week
2010-12-09 02:42:02 +00:00
if (td->td_lend_user_pri > pri)
sched_lend_user_prio(td, pri);
else {
thread_unlock(td);
break;
}
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_unlock(td);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Pick up the lock that td is blocked on.
*/
uq = td->td_umtxq;
pi = uq->uq_pi_blocked;
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
if (pi == NULL)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
break;
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
/* Resort td on the list if needed. */
umtx_pi_adjust_thread(pi, td);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
}
/*
* Unpropagate priority for a PI mutex when a thread blocked on
* it is interrupted by signal or resumed by others.
*/
static void
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
umtx_repropagate_priority(struct umtx_pi *pi)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
struct umtx_q *uq, *uq_owner;
struct umtx_pi *pi2;
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
int pri;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
mtx_assert(&umtx_lock, MA_OWNED);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
The dependency chain for priority-inheritance mutexes could be subverted by userspace into cycle. Both umtx_propagate_priority() and umtx_repropagate_priority() would then loop infinitely, owning the spinlock. Check for the cycle using standard Floyd' algorithm before doing the pass in the affected functions. Add simple check for condition of tricking the thread into a wait for itself, which could be easily simulated by usermode without race. Found by: Eric van Gyzen <eric@vangyzen.net> In collaboration with: Eric van Gyzen <eric@vangyzen.net> Tested by: pho MFC after: 1 week
2015-01-31 12:27:40 +00:00
if (umtx_pi_check_loop(pi))
return;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
while (pi != NULL && pi->pi_owner != NULL) {
pri = PRI_MAX;
uq_owner = pi->pi_owner->td_umtxq;
TAILQ_FOREACH(pi2, &uq_owner->uq_pi_contested, pi_link) {
uq = TAILQ_FIRST(&pi2->pi_blocked);
if (uq != NULL) {
if (pri > UPRI(uq->uq_thread))
pri = UPRI(uq->uq_thread);
}
}
if (pri > uq_owner->uq_inherited_pri)
pri = uq_owner->uq_inherited_pri;
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_lock(pi->pi_owner);
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
sched_lend_user_prio(pi->pi_owner, pri);
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_unlock(pi->pi_owner);
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
if ((pi = uq_owner->uq_pi_blocked) != NULL)
umtx_pi_adjust_thread(pi, uq_owner->uq_thread);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
}
/*
* Insert a PI mutex into owned list.
*/
static void
umtx_pi_setowner(struct umtx_pi *pi, struct thread *owner)
{
struct umtx_q *uq_owner;
uq_owner = owner->td_umtxq;
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
mtx_assert(&umtx_lock, MA_OWNED);
Convert explicit panic() call to assert. Based on github pull request: #113 Submitted by: pmarillo@github MFC after: 1 week
2017-11-04 10:49:34 +00:00
MPASS(pi->pi_owner == NULL);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
pi->pi_owner = owner;
TAILQ_INSERT_TAIL(&uq_owner->uq_pi_contested, pi, pi_link);
}
When unlocking a contested PI pthread mutex, if the queue of waiters is empty, look up the umtx_pi and disown it if the current thread owns it. This can happen if a signal or timeout removed the last waiter from the queue, but there is still a thread in do_lock_pi() holding a reference on the umtx_pi. The unlocking thread might not own the umtx_pi in this case, but if it does, it must disown it to keep the ownership consistent between the umtx_pi and the umutex. Submitted by: Eric van Gyzen <eric_van_gyzen@dell.com> with advice from: Elliott Rabe and Jim Muchow, also at Dell Inc. Obtained from: Dell Inc. PR: 198914
2015-02-25 16:12:56 +00:00
/*
* Disown a PI mutex, and remove it from the owned list.
*/
static void
umtx_pi_disown(struct umtx_pi *pi)
{
mtx_assert(&umtx_lock, MA_OWNED);
TAILQ_REMOVE(&pi->pi_owner->td_umtxq->uq_pi_contested, pi, pi_link);
pi->pi_owner = NULL;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Claim ownership of a PI mutex.
*/
static int
umtx_pi_claim(struct umtx_pi *pi, struct thread *owner)
{
Clean up some cosmetic nits in kern_umtx.c, found during recent work in this area and by the Clang static analyzer. Remove some dead assignments. Fix a typo in a panic string. Use umtx_pi_disown() instead of duplicate code. Use an existing variable instead of curthread. Approved by: kib (mentor) MFC after: 3 days Sponsored by: Dell Inc
2015-03-28 21:21:40 +00:00
struct umtx_q *uq;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
int pri;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (pi->pi_owner == owner) {
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (0);
}
if (pi->pi_owner != NULL) {
/*
* userland may have already messed the mutex, sigh.
*/
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EPERM);
}
umtx_pi_setowner(pi, owner);
uq = TAILQ_FIRST(&pi->pi_blocked);
if (uq != NULL) {
pri = UPRI(uq->uq_thread);
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_lock(owner);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (pri < UPRI(owner))
sched_lend_user_prio(owner, pri);
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_unlock(owner);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (0);
}
Add missing changes for fixing LOR of umtx lock and thread lock, follow the committing of files: kern_resource.c revision 1.181 sched_4bsd.c revision 1.111 sched_ule.c revision 1.218
2007-12-17 05:55:07 +00:00
/*
* Adjust a thread's order position in its blocked PI mutex,
* this may result new priority propagating process.
*/
void
umtx_pi_adjust(struct thread *td, u_char oldpri)
{
Check NULL pointer.
2007-12-17 08:09:37 +00:00
struct umtx_q *uq;
struct umtx_pi *pi;
uq = td->td_umtxq;
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
Check NULL pointer.
2007-12-17 08:09:37 +00:00
/*
* Pick up the lock that td is blocked on.
*/
pi = uq->uq_pi_blocked;
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
if (pi != NULL) {
umtx_pi_adjust_thread(pi, td);
umtx_repropagate_priority(pi);
}
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
Add missing changes for fixing LOR of umtx lock and thread lock, follow the committing of files: kern_resource.c revision 1.181 sched_4bsd.c revision 1.111 sched_ule.c revision 1.218
2007-12-17 05:55:07 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Sleep on a PI mutex.
*/
static int
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
umtxq_sleep_pi(struct umtx_q *uq, struct umtx_pi *pi, uint32_t owner,
const char *wmesg, struct abs_timeout *timo, bool shared)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
struct thread *td, *td1;
struct umtx_q *uq1;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
int error, pri;
umtx: don't call umtxq_getchain unless the value is needed
2018-05-19 05:09:10 +00:00
#ifdef INVARIANTS
struct umtxq_chain *uc;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx: don't call umtxq_getchain unless the value is needed
2018-05-19 05:09:10 +00:00
uc = umtxq_getchain(&pi->pi_key);
#endif
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
error = 0;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
td = uq->uq_thread;
KASSERT(td == curthread, ("inconsistent uq_thread"));
umtx: don't call umtxq_getchain unless the value is needed
2018-05-19 05:09:10 +00:00
UMTXQ_LOCKED_ASSERT(umtxq_getchain(&uq->uq_key));
Fix assertion, &uc->uc_busy is never zero, the intent is to test the uc_busy value, and not its address [1]. Remove the single use of the macro, write KASSERT() explicitely in the code of umtxq_sleep_pi(). Submitted by: Eric van Gyzen <eric@vangyzen.net> [1] MFC after: 1 week
2014-11-13 18:51:09 +00:00
KASSERT(uc->uc_busy != 0, ("umtx chain is not busy"));
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_insert(uq);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (pi->pi_owner == NULL) {
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
td1 = tdfind(owner, shared ? -1 : td->td_proc->p_pid);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
Only unlock process if a thread is found.
2010-11-15 07:33:54 +00:00
if (td1 != NULL) {
if (pi->pi_owner == NULL)
umtx_pi_setowner(pi, td1);
PROC_UNLOCK(td1->td_proc);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
}
TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) {
pri = UPRI(uq1->uq_thread);
if (pri > UPRI(td))
break;
}
if (uq1 != NULL)
TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq);
else
TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq);
uq->uq_pi_blocked = pi;
Add missing changes for fixing LOR of umtx lock and thread lock, follow the committing of files: kern_resource.c revision 1.181 sched_4bsd.c revision 1.111 sched_ule.c revision 1.218
2007-12-17 05:55:07 +00:00
thread_lock(td);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
td->td_flags |= TDF_UPIBLOCKED;
Add missing changes for fixing LOR of umtx lock and thread lock, follow the committing of files: kern_resource.c revision 1.181 sched_4bsd.c revision 1.111 sched_ule.c revision 1.218
2007-12-17 05:55:07 +00:00
thread_unlock(td);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx_propagate_priority(td);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
umtxq_unbusy(&uq->uq_key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = umtxq_sleep(uq, wmesg, timo);
umtxq_remove(uq);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
uq->uq_pi_blocked = NULL;
Add missing changes for fixing LOR of umtx lock and thread lock, follow the committing of files: kern_resource.c revision 1.181 sched_4bsd.c revision 1.111 sched_ule.c revision 1.218
2007-12-17 05:55:07 +00:00
thread_lock(td);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
td->td_flags &= ~TDF_UPIBLOCKED;
Add missing changes for fixing LOR of umtx lock and thread lock, follow the committing of files: kern_resource.c revision 1.181 sched_4bsd.c revision 1.111 sched_ule.c revision 1.218
2007-12-17 05:55:07 +00:00
thread_unlock(td);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq);
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
umtx_repropagate_priority(pi);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
umtxq_unlock(&uq->uq_key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (error);
}
/*
* Add reference count for a PI mutex.
*/
static void
umtx_pi_ref(struct umtx_pi *pi)
{
umtx: don't call umtxq_getchain unless the value is needed
2018-05-19 05:09:10 +00:00
UMTXQ_LOCKED_ASSERT(umtxq_getchain(&pi->pi_key));
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
pi->pi_refcount++;
}
/*
* Decrease reference count for a PI mutex, if the counter
* is decreased to zero, its memory space is freed.
*/
static void
umtx_pi_unref(struct umtx_pi *pi)
{
struct umtxq_chain *uc;
uc = umtxq_getchain(&pi->pi_key);
UMTXQ_LOCKED_ASSERT(uc);
KASSERT(pi->pi_refcount > 0, ("invalid reference count"));
if (--pi->pi_refcount == 0) {
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
Clean up some cosmetic nits in kern_umtx.c, found during recent work in this area and by the Clang static analyzer. Remove some dead assignments. Fix a typo in a panic string. Use umtx_pi_disown() instead of duplicate code. Use an existing variable instead of curthread. Approved by: kib (mentor) MFC after: 3 days Sponsored by: Dell Inc
2015-03-28 21:21:40 +00:00
if (pi->pi_owner != NULL)
umtx_pi_disown(pi);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
KASSERT(TAILQ_EMPTY(&pi->pi_blocked),
("blocked queue not empty"));
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
TAILQ_REMOVE(&uc->uc_pi_list, pi, pi_hashlink);
umtx_pi_free(pi);
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
/*
* Find a PI mutex in hash table.
*/
static struct umtx_pi *
umtx_pi_lookup(struct umtx_key *key)
{
struct umtxq_chain *uc;
struct umtx_pi *pi;
uc = umtxq_getchain(key);
UMTXQ_LOCKED_ASSERT(uc);
TAILQ_FOREACH(pi, &uc->uc_pi_list, pi_hashlink) {
if (umtx_key_match(&pi->pi_key, key)) {
return (pi);
}
}
return (NULL);
}
/*
* Insert a PI mutex into hash table.
*/
static inline void
umtx_pi_insert(struct umtx_pi *pi)
{
struct umtxq_chain *uc;
uc = umtxq_getchain(&pi->pi_key);
UMTXQ_LOCKED_ASSERT(uc);
TAILQ_INSERT_TAIL(&uc->uc_pi_list, pi, pi_hashlink);
}
/*
* Lock a PI mutex.
*/
static int
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
do_lock_pi(struct thread *td, struct umutex *m, uint32_t flags,
struct _umtx_time *timeout, int try)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout timo;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtx_q *uq;
struct umtx_pi *pi, *new_pi;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uint32_t id, old_owner, owner, old;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
int error, rv;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
id = td->td_tid;
uq = td->td_umtxq;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
TYPE_PI_ROBUST_UMUTEX : TYPE_PI_UMUTEX, GET_SHARE(flags),
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
&uq->uq_key)) != 0)
return (error);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
if (timeout != NULL)
abs_timeout_init2(&timo, timeout);
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
umtxq_lock(&uq->uq_key);
pi = umtx_pi_lookup(&uq->uq_key);
if (pi == NULL) {
new_pi = umtx_pi_alloc(M_NOWAIT);
if (new_pi == NULL) {
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_unlock(&uq->uq_key);
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
new_pi = umtx_pi_alloc(M_WAITOK);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_lock(&uq->uq_key);
pi = umtx_pi_lookup(&uq->uq_key);
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
if (pi != NULL) {
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx_pi_free(new_pi);
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
new_pi = NULL;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
}
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
if (new_pi != NULL) {
new_pi->pi_key = uq->uq_key;
umtx_pi_insert(new_pi);
pi = new_pi;
}
}
umtx_pi_ref(pi);
umtxq_unlock(&uq->uq_key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
/*
* Care must be exercised when dealing with umtx structure. It
* can fault on any access.
*/
for (;;) {
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Try the uncontested case. This should be done in userland.
*/
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&m->m_owner, UMUTEX_UNOWNED, &owner, id);
/* The address was invalid. */
if (rv == -1) {
error = EFAULT;
break;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* The acquire succeeded. */
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(owner == UMUTEX_UNOWNED);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
error = 0;
break;
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (owner == UMUTEX_RB_NOTRECOV) {
error = ENOTRECOVERABLE;
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
/*
* Avoid overwriting a possible error from sleep due
* to the pending signal with suspension check result.
*/
if (error == 0) {
error = umtxq_check_susp(td, true);
if (error != 0)
break;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* If no one owns it but it is contested try to acquire it. */
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (owner == UMUTEX_CONTESTED || owner == UMUTEX_RB_OWNERDEAD) {
old_owner = owner;
rv = casueword32(&m->m_owner, owner, &owner,
id | UMUTEX_CONTESTED);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
/* The address was invalid. */
if (rv == -1) {
error = EFAULT;
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 1) {
if (error == 0) {
error = umtxq_check_susp(td, true);
if (error != 0)
In do_lock_pi(), do not return prematurely. If umtxq_check_susp() indicates an exit, we should clean the resources before returning. Do it by breaking out of the loop and relying on post-loop cleanup. Reviewed by: markj Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 12 days Differential revision: https://reviews.freebsd.org/D20949
2019-07-15 08:39:52 +00:00
break;
When failing to claim ownership of a umtx_pi, restore the umutex owner to its previous, unowned state. This avoids compounding an existing problem of inconsistent ownership. Submitted by: Eric van Gyzen <eric_van_gyzen@dell.com> Obtained from: Dell Inc. PR: 198914 MFC after: 1 week
2015-02-25 16:17:16 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
/*
* If this failed the lock could
* changed, restart.
*/
continue;
}
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
MPASS(rv == 0);
MPASS(owner == old_owner);
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
error = umtx_pi_claim(pi, td);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
if (error != 0) {
/*
* Since we're going to return an
* error, restore the m_owner to its
* previous, unowned state to avoid
* compounding the problem.
*/
(void)casuword32(&m->m_owner,
id | UMUTEX_CONTESTED, old_owner);
}
if (error == 0 && old_owner == UMUTEX_RB_OWNERDEAD)
error = EOWNERDEAD;
break;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
The dependency chain for priority-inheritance mutexes could be subverted by userspace into cycle. Both umtx_propagate_priority() and umtx_repropagate_priority() would then loop infinitely, owning the spinlock. Check for the cycle using standard Floyd' algorithm before doing the pass in the affected functions. Add simple check for condition of tricking the thread into a wait for itself, which could be easily simulated by usermode without race. Found by: Eric van Gyzen <eric@vangyzen.net> In collaboration with: Eric van Gyzen <eric@vangyzen.net> Tested by: pho MFC after: 1 week
2015-01-31 12:27:40 +00:00
if ((owner & ~UMUTEX_CONTESTED) == id) {
error = EDEADLK;
break;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (try != 0) {
error = EBUSY;
break;
}
/*
* If we caught a signal, we have retried and now
* exit immediately.
*/
if (error != 0)
break;
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
/*
* Set the contested bit so that a release in user space
* knows to use the system call for unlock. If this fails
* either some one else has acquired the lock or it has been
* released.
*/
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
rv = casueword32(&m->m_owner, owner, &old, owner |
UMUTEX_CONTESTED);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* The address was invalid. */
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1) {
umtxq_unbusy_unlocked(&uq->uq_key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
error = EFAULT;
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 1) {
umtxq_unbusy_unlocked(&uq->uq_key);
error = umtxq_check_susp(td, true);
In do_lock_pi(), do not override error from umtxq_sleep_pi() when doing suspend check. This restores the pre-r251684 behaviour, to retry once after the signal is detected. PR: kern/192918 Submitted by: Elliott Rabe, Dell Inc., Eric van Gyzen <eric@vangyzen.net> Obtained from: Dell Inc. MFC after: 1 week
2014-08-22 18:42:14 +00:00
if (error != 0)
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
break;
/*
* The lock changed and we need to retry or we
* lost a race to the thread unlocking the
* umtx. Note that the UMUTEX_RB_OWNERDEAD
* value for owner is impossible there.
*/
continue;
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
}
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
umtxq_lock(&uq->uq_key);
/* We set the contested bit, sleep. */
MPASS(old == owner);
error = umtxq_sleep_pi(uq, pi, owner & ~UMUTEX_CONTESTED,
"umtxpi", timeout == NULL ? NULL : &timo,
(flags & USYNC_PROCESS_SHARED) != 0);
if (error != 0)
continue;
error = umtxq_check_susp(td, false);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
break;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
Optimize umtx_lock_pi() a bit by moving some heavy code out of the loop, make a fast path when a umtx_pi can be allocated without being blocked.
2006-10-26 09:33:34 +00:00
umtxq_lock(&uq->uq_key);
umtx_pi_unref(pi);
umtxq_unlock(&uq->uq_key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx_key_release(&uq->uq_key);
return (error);
}
/*
* Unlock a PI mutex.
*/
static int
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
do_unlock_pi(struct thread *td, struct umutex *m, uint32_t flags, bool rb)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
struct umtx_key key;
struct umtx_q *uq_first, *uq_first2, *uq_me;
struct umtx_pi *pi, *pi2;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uint32_t id, new_owner, old, owner;
int count, error, pri;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
id = td->td_tid;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
usrloop:
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Make sure we own this mtx.
*/
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = fueword32(&m->m_owner, &owner);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
if ((owner & ~UMUTEX_CONTESTED) != id)
return (EPERM);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
new_owner = umtx_unlock_val(flags, rb);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* This should be done in userland */
if ((owner & UMUTEX_CONTESTED) == 0) {
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
error = casueword32(&m->m_owner, owner, &old, new_owner);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (error == 1) {
error = umtxq_check_susp(td, true);
if (error != 0)
return (error);
goto usrloop;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (old == owner)
return (0);
Make sure we get new m_owner value if we can not unlock it in uncontested case. Reorder statements in do_unlock_umutex.
2006-09-02 02:41:33 +00:00
owner = old;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
/* We should only ever be in here for contested locks */
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
TYPE_PI_ROBUST_UMUTEX : TYPE_PI_UMUTEX, GET_SHARE(flags),
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
&key)) != 0)
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
count = umtxq_count_pi(&key, &uq_first);
if (uq_first != NULL) {
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
pi = uq_first->uq_pi_blocked;
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
KASSERT(pi != NULL, ("pi == NULL?"));
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (pi->pi_owner != td && !(rb && pi->pi_owner == NULL)) {
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_unbusy(&key);
umtxq_unlock(&key);
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
umtx_key_release(&key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/* userland messed the mutex */
return (EPERM);
}
Clean up some cosmetic nits in kern_umtx.c, found during recent work in this area and by the Clang static analyzer. Remove some dead assignments. Fix a typo in a panic string. Use umtx_pi_disown() instead of duplicate code. Use an existing variable instead of curthread. Approved by: kib (mentor) MFC after: 3 days Sponsored by: Dell Inc
2015-03-28 21:21:40 +00:00
uq_me = td->td_umtxq;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (pi->pi_owner == td)
umtx_pi_disown(pi);
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
/* get highest priority thread which is still sleeping. */
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
uq_first = TAILQ_FIRST(&pi->pi_blocked);
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
while (uq_first != NULL &&
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
(uq_first->uq_flags & UQF_UMTXQ) == 0) {
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
uq_first = TAILQ_NEXT(uq_first, uq_lockq);
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
pri = PRI_MAX;
TAILQ_FOREACH(pi2, &uq_me->uq_pi_contested, pi_link) {
uq_first2 = TAILQ_FIRST(&pi2->pi_blocked);
if (uq_first2 != NULL) {
if (pri > UPRI(uq_first2->uq_thread))
pri = UPRI(uq_first2->uq_thread);
}
}
Clean up some cosmetic nits in kern_umtx.c, found during recent work in this area and by the Clang static analyzer. Remove some dead assignments. Fix a typo in a panic string. Use umtx_pi_disown() instead of duplicate code. Use an existing variable instead of curthread. Approved by: kib (mentor) MFC after: 3 days Sponsored by: Dell Inc
2015-03-28 21:21:40 +00:00
thread_lock(td);
sched_lend_user_prio(td, pri);
thread_unlock(td);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
1) Check NULL pointer before calling umtx_pi_adjust_locked(), this avoids a PANIC. 2) Rework locking for POSIX priority-mutex, this fixes a race where a thread may wait there forever even if the mutex is unlocked.
2009-03-13 06:06:20 +00:00
if (uq_first)
umtxq_signal_thread(uq_first);
When unlocking a contested PI pthread mutex, if the queue of waiters is empty, look up the umtx_pi and disown it if the current thread owns it. This can happen if a signal or timeout removed the last waiter from the queue, but there is still a thread in do_lock_pi() holding a reference on the umtx_pi. The unlocking thread might not own the umtx_pi in this case, but if it does, it must disown it to keep the ownership consistent between the umtx_pi and the umutex. Submitted by: Eric van Gyzen <eric_van_gyzen@dell.com> with advice from: Elliott Rabe and Jim Muchow, also at Dell Inc. Obtained from: Dell Inc. PR: 198914
2015-02-25 16:12:56 +00:00
} else {
pi = umtx_pi_lookup(&key);
/*
* A umtx_pi can exist if a signal or timeout removed the
* last waiter from the umtxq, but there is still
* a thread in do_lock_pi() holding the umtx_pi.
*/
if (pi != NULL) {
/*
* The umtx_pi can be unowned, such as when a thread
* has just entered do_lock_pi(), allocated the
* umtx_pi, and unlocked the umtxq.
* If the current thread owns it, it must disown it.
*/
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
When unlocking a contested PI pthread mutex, if the queue of waiters is empty, look up the umtx_pi and disown it if the current thread owns it. This can happen if a signal or timeout removed the last waiter from the queue, but there is still a thread in do_lock_pi() holding a reference on the umtx_pi. The unlocking thread might not own the umtx_pi in this case, but if it does, it must disown it to keep the ownership consistent between the umtx_pi and the umutex. Submitted by: Eric van Gyzen <eric_van_gyzen@dell.com> with advice from: Elliott Rabe and Jim Muchow, also at Dell Inc. Obtained from: Dell Inc. PR: 198914
2015-02-25 16:12:56 +00:00
if (pi->pi_owner == td)
umtx_pi_disown(pi);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
When unlocking a contested PI pthread mutex, if the queue of waiters is empty, look up the umtx_pi and disown it if the current thread owns it. This can happen if a signal or timeout removed the last waiter from the queue, but there is still a thread in do_lock_pi() holding a reference on the umtx_pi. The unlocking thread might not own the umtx_pi in this case, but if it does, it must disown it to keep the ownership consistent between the umtx_pi and the umutex. Submitted by: Eric van Gyzen <eric_van_gyzen@dell.com> with advice from: Elliott Rabe and Jim Muchow, also at Dell Inc. Obtained from: Dell Inc. PR: 198914
2015-02-25 16:12:56 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
umtxq_unlock(&key);
/*
* When unlocking the umtx, it must be marked as unowned if
* there is zero or one thread only waiting for it.
* Otherwise, it must be marked as contested.
*/
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (count > 1)
new_owner |= UMUTEX_CONTESTED;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
again:
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
error = casueword32(&m->m_owner, owner, &old, new_owner);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (error == 1) {
error = umtxq_check_susp(td, false);
if (error == 0)
goto again;
}
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx_key_release(&key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (error == 0 && old != owner)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EINVAL);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
return (error);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
/*
* Lock a PP mutex.
*/
static int
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
do_lock_pp(struct thread *td, struct umutex *m, uint32_t flags,
struct _umtx_time *timeout, int try)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout timo;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
struct umtx_q *uq, *uq2;
struct umtx_pi *pi;
uint32_t ceiling;
uint32_t owner, id;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
int error, pri, old_inherited_pri, su, rv;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
id = td->td_tid;
uq = td->td_umtxq;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
TYPE_PP_ROBUST_UMUTEX : TYPE_PP_UMUTEX, GET_SHARE(flags),
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
&uq->uq_key)) != 0)
return (error);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
if (timeout != NULL)
abs_timeout_init2(&timo, timeout);
Sweep kernel replacing suser(9) calls with priv(9) calls, assigning specific privilege names to a broad range of privileges. These may require some future tweaking. Sponsored by: nCircle Network Security, Inc. Obtained from: TrustedBSD Project Discussed on: arch@ Reviewed (at least in part) by: mlaier, jmg, pjd, bde, ceri, Alex Lyashkov <umka at sevcity dot net>, Skip Ford <skip dot ford at verizon dot net>, Antoine Brodin <antoine dot brodin at laposte dot net>
2006-11-06 13:42:10 +00:00
su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
for (;;) {
old_inherited_pri = uq->uq_inherited_pri;
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = fueword32(&m->m_ceilings[0], &ceiling);
if (rv == -1) {
error = EFAULT;
goto out;
}
ceiling = RTP_PRIO_MAX - ceiling;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (ceiling > RTP_PRIO_MAX) {
error = EINVAL;
goto out;
}
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (UPRI(td) < PRI_MIN_REALTIME + ceiling) {
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
error = EINVAL;
goto out;
}
if (su && PRI_MIN_REALTIME + ceiling < uq->uq_inherited_pri) {
uq->uq_inherited_pri = PRI_MIN_REALTIME + ceiling;
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_lock(td);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if (uq->uq_inherited_pri < UPRI(td))
sched_lend_user_prio(td, uq->uq_inherited_pri);
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_unlock(td);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
rv = casueword32(&m->m_owner, UMUTEX_CONTESTED, &owner,
id | UMUTEX_CONTESTED);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
/* The address was invalid. */
if (rv == -1) {
error = EFAULT;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(owner == UMUTEX_CONTESTED);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = 0;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
break;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
/* rv == 1 */
if (owner == UMUTEX_RB_OWNERDEAD) {
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
rv = casueword32(&m->m_owner, UMUTEX_RB_OWNERDEAD,
&owner, id | UMUTEX_CONTESTED);
if (rv == -1) {
error = EFAULT;
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(owner == UMUTEX_RB_OWNERDEAD);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
error = EOWNERDEAD; /* success */
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
/*
* rv == 1, only check for suspension if we
* did not already catched a signal. If we
* get an error from the check, the same
* condition is checked by the umtxq_sleep()
* call below, so we should obliterate the
* error to not skip the last loop iteration.
*/
if (error == 0) {
error = umtxq_check_susp(td, false);
if (error == 0) {
if (try != 0)
error = EBUSY;
else
continue;
}
error = 0;
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
} else if (owner == UMUTEX_RB_NOTRECOV) {
error = ENOTRECOVERABLE;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (try != 0)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
error = EBUSY;
/*
* If we caught a signal, we have retried and now
* exit immediately.
*/
if (error != 0)
break;
umtxq_lock(&uq->uq_key);
umtxq_insert(uq);
umtxq_unbusy(&uq->uq_key);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = umtxq_sleep(uq, "umtxpp", timeout == NULL ?
NULL : &timo);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
uq->uq_inherited_pri = old_inherited_pri;
pri = PRI_MAX;
TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
uq2 = TAILQ_FIRST(&pi->pi_blocked);
if (uq2 != NULL) {
if (pri > UPRI(uq2->uq_thread))
pri = UPRI(uq2->uq_thread);
}
}
if (pri > uq->uq_inherited_pri)
pri = uq->uq_inherited_pri;
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_lock(td);
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
sched_lend_user_prio(td, pri);
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_unlock(td);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (error != 0 && error != EOWNERDEAD) {
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
uq->uq_inherited_pri = old_inherited_pri;
pri = PRI_MAX;
TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
uq2 = TAILQ_FIRST(&pi->pi_blocked);
if (uq2 != NULL) {
if (pri > UPRI(uq2->uq_thread))
pri = UPRI(uq2->uq_thread);
}
}
if (pri > uq->uq_inherited_pri)
pri = uq->uq_inherited_pri;
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_lock(td);
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
sched_lend_user_prio(td, pri);
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_unlock(td);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
out:
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&uq->uq_key);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx_key_release(&uq->uq_key);
return (error);
}
/*
* Unlock a PP mutex.
*/
static int
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags, bool rb)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
struct umtx_key key;
struct umtx_q *uq, *uq2;
struct umtx_pi *pi;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uint32_t id, owner, rceiling;
Check if it is root user in do_unlock_pp.
2006-09-03 00:07:37 +00:00
int error, pri, new_inherited_pri, su;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
id = td->td_tid;
uq = td->td_umtxq;
Sweep kernel replacing suser(9) calls with priv(9) calls, assigning specific privilege names to a broad range of privileges. These may require some future tweaking. Sponsored by: nCircle Network Security, Inc. Obtained from: TrustedBSD Project Discussed on: arch@ Reviewed (at least in part) by: mlaier, jmg, pjd, bde, ceri, Alex Lyashkov <umka at sevcity dot net>, Skip Ford <skip dot ford at verizon dot net>, Antoine Brodin <antoine dot brodin at laposte dot net>
2006-11-06 13:42:10 +00:00
su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
* Make sure we own this mtx.
*/
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = fueword32(&m->m_owner, &owner);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
if ((owner & ~UMUTEX_CONTESTED) != id)
return (EPERM);
error = copyin(&m->m_ceilings[1], &rceiling, sizeof(uint32_t));
if (error != 0)
return (error);
if (rceiling == -1)
new_inherited_pri = PRI_MAX;
else {
rceiling = RTP_PRIO_MAX - rceiling;
if (rceiling > RTP_PRIO_MAX)
return (EINVAL);
new_inherited_pri = PRI_MIN_REALTIME + rceiling;
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
TYPE_PP_ROBUST_UMUTEX : TYPE_PP_UMUTEX, GET_SHARE(flags),
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
&key)) != 0)
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
umtxq_unlock(&key);
/*
* For priority protected mutex, always set unlocked state
* to UMUTEX_CONTESTED, so that userland always enters kernel
* to lock the mutex, it is necessary because thread priority
* has to be adjusted for such mutex.
*/
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
error = suword32(&m->m_owner, umtx_unlock_val(flags, rb) |
UMUTEX_CONTESTED);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_lock(&key);
if (error == 0)
umtxq_signal(&key, 1);
umtxq_unbusy(&key);
umtxq_unlock(&key);
if (error == -1)
error = EFAULT;
else {
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_lock(&umtx_lock);
Check if it is root user in do_unlock_pp.
2006-09-03 00:07:37 +00:00
if (su != 0)
uq->uq_inherited_pri = new_inherited_pri;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
pri = PRI_MAX;
TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
uq2 = TAILQ_FIRST(&pi->pi_blocked);
if (uq2 != NULL) {
if (pri > UPRI(uq2->uq_thread))
pri = UPRI(uq2->uq_thread);
}
}
if (pri > uq->uq_inherited_pri)
pri = uq->uq_inherited_pri;
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_lock(td);
- Follow r216313, the sched_unlend_user_prio is no longer needed, always use sched_lend_user_prio to set lent priority. - Improve pthread priority-inherit mutex, when a contender's priority is lowered, repropagete priorities, this may cause mutex owner's priority to be lowerd, in old code, mutex owner's priority is rise-only.
2010-12-29 09:26:46 +00:00
sched_lend_user_prio(td, pri);
Commit 8/14 of sched_lock decomposition. - Use a global umtx spinlock to protect the sleep queues now that there is no global scheduler lock. - Use thread_lock() to protect thread state. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:54:50 +00:00
thread_unlock(td);
The umtx_lock mutex is used by top-half of the kernel, but is currently a spin lock. Apparently, the only reason for this is that umtx_thread_exit() is called under the process spinlock, which put the requirement on the umtx_lock. Note that the witness static order list is wrong for the umtx_lock, umtx_lock is explicitely before any thread lock, so it is also before sleepq locks. Change umtx_lock to be the sleepable mutex. For the reason above, the calls to umtx_thread_exit() are moved from thread_exit() earlier in each caller, when the process spin lock is not yet taken. Discussed with: jhb Tested by: pho (previous version) Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2015-02-28 04:19:02 +00:00
mtx_unlock(&umtx_lock);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
umtx_key_release(&key);
return (error);
}
static int
do_set_ceiling(struct thread *td, struct umutex *m, uint32_t ceiling,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uint32_t *old_ceiling)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
struct umtx_q *uq;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uint32_t flags, id, owner, save_ceiling;
int error, rv, rv1;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&m->m_flags, &flags);
if (error == -1)
return (EFAULT);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
if ((flags & UMUTEX_PRIO_PROTECT) == 0)
return (EINVAL);
if (ceiling > RTP_PRIO_MAX)
return (EINVAL);
id = td->td_tid;
uq = td->td_umtxq;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
TYPE_PP_ROBUST_UMUTEX : TYPE_PP_UMUTEX, GET_SHARE(flags),
&uq->uq_key)) != 0)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (error);
for (;;) {
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = fueword32(&m->m_ceilings[0], &save_ceiling);
if (rv == -1) {
error = EFAULT;
break;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
rv = casueword32(&m->m_owner, UMUTEX_CONTESTED, &owner,
id | UMUTEX_CONTESTED);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1) {
error = EFAULT;
break;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Correctly check for casueword(9) success in do_set_ceiling(). After r349951, the return code must be checked instead of old == new comparision. Reviewed by: markj Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 12 days Differential revision: https://reviews.freebsd.org/D20949
2019-07-15 08:38:01 +00:00
if (rv == 0) {
MPASS(owner == UMUTEX_CONTESTED);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
rv = suword32(&m->m_ceilings[0], ceiling);
rv1 = suword32(&m->m_owner, UMUTEX_CONTESTED);
error = (rv == 0 && rv1 == 0) ? 0: EFAULT;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
break;
}
if ((owner & ~UMUTEX_CONTESTED) == id) {
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
rv = suword32(&m->m_ceilings[0], ceiling);
error = rv == 0 ? 0 : EFAULT;
break;
}
if (owner == UMUTEX_RB_OWNERDEAD) {
error = EOWNERDEAD;
break;
} else if (owner == UMUTEX_RB_NOTRECOV) {
error = ENOTRECOVERABLE;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
break;
}
/*
* If we caught a signal, we have retried and now
* exit immediately.
*/
if (error != 0)
break;
/*
* We set the contested bit, sleep. Otherwise the lock changed
* and we need to retry or we lost a race to the thread
* unlocking the umtx.
*/
umtxq_lock(&uq->uq_key);
umtxq_insert(uq);
umtxq_unbusy(&uq->uq_key);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = umtxq_sleep(uq, "umtxpp", NULL);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
}
umtxq_lock(&uq->uq_key);
if (error == 0)
umtxq_signal(&uq->uq_key, INT_MAX);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
if (error == 0 && old_ceiling != NULL) {
rv = suword32(old_ceiling, save_ceiling);
error = rv == 0 ? 0 : EFAULT;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (error);
}
/*
* Lock a userland POSIX mutex.
*/
static int
Merge all code of do_lock_normal, do_lock_pi and do_lock_pp into function do_lock_umutex.
2006-09-05 12:01:09 +00:00
do_lock_umutex(struct thread *td, struct umutex *m,
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct _umtx_time *timeout, int mode)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
uint32_t flags;
Merge all code of do_lock_normal, do_lock_pi and do_lock_pp into function do_lock_umutex.
2006-09-05 12:01:09 +00:00
int error;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&m->m_flags, &flags);
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
switch (flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
case 0:
error = do_lock_normal(td, m, flags, timeout, mode);
break;
case UMUTEX_PRIO_INHERIT:
error = do_lock_pi(td, m, flags, timeout, mode);
break;
case UMUTEX_PRIO_PROTECT:
error = do_lock_pp(td, m, flags, timeout, mode);
break;
default:
return (EINVAL);
}
Merge all code of do_lock_normal, do_lock_pi and do_lock_pp into function do_lock_umutex.
2006-09-05 12:01:09 +00:00
if (timeout == NULL) {
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
if (error == EINTR && mode != _UMUTEX_WAIT)
Merge all code of do_lock_normal, do_lock_pi and do_lock_pp into function do_lock_umutex.
2006-09-05 12:01:09 +00:00
error = ERESTART;
} else {
/* Timed-locking is not restarted. */
if (error == ERESTART)
error = EINTR;
Reorder some statments. Fix typo and remove stale comments.
2006-08-30 23:59:45 +00:00
}
Merge all code of do_lock_normal, do_lock_pi and do_lock_pp into function do_lock_umutex.
2006-09-05 12:01:09 +00:00
return (error);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
/*
* Unlock a userland POSIX mutex.
*/
static int
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
do_unlock_umutex(struct thread *td, struct umutex *m, bool rb)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
uint32_t flags;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
int error;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&m->m_flags, &flags);
if (error == -1)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
return (EFAULT);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
switch (flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
Make sure we get new m_owner value if we can not unlock it in uncontested case. Reorder statements in do_unlock_umutex.
2006-09-02 02:41:33 +00:00
case 0:
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_unlock_normal(td, m, flags, rb));
Make sure we get new m_owner value if we can not unlock it in uncontested case. Reorder statements in do_unlock_umutex.
2006-09-02 02:41:33 +00:00
case UMUTEX_PRIO_INHERIT:
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_unlock_pi(td, m, flags, rb));
Make sure we get new m_owner value if we can not unlock it in uncontested case. Reorder statements in do_unlock_umutex.
2006-09-02 02:41:33 +00:00
case UMUTEX_PRIO_PROTECT:
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_unlock_pp(td, m, flags, rb));
Make sure we get new m_owner value if we can not unlock it in uncontested case. Reorder statements in do_unlock_umutex.
2006-09-02 02:41:33 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Make sure we get new m_owner value if we can not unlock it in uncontested case. Reorder statements in do_unlock_umutex.
2006-09-02 02:41:33 +00:00
return (EINVAL);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
static int
do_cv_wait(struct thread *td, struct ucond *cv, struct umutex *m,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
struct timespec *timeout, u_long wflags)
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
{
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout timo;
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
struct umtx_q *uq;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
uint32_t flags, clockid, hasw;
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
int error;
uq = td->td_umtxq;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&cv->c_flags, &flags);
if (error == -1)
return (EFAULT);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &uq->uq_key);
if (error != 0)
return (error);
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
if ((wflags & CVWAIT_CLOCKID) != 0) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&cv->c_clockid, &clockid);
if (error == -1) {
umtx_key_release(&uq->uq_key);
return (EFAULT);
}
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
if (clockid < CLOCK_REALTIME ||
clockid >= CLOCK_THREAD_CPUTIME_ID) {
/* hmm, only HW clock id will work. */
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtx_key_release(&uq->uq_key);
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
return (EINVAL);
}
} else {
clockid = CLOCK_REALTIME;
}
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_insert(uq);
umtxq_unlock(&uq->uq_key);
/*
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
* Set c_has_waiters to 1 before releasing user mutex, also
* don't modify cache line when unnecessary.
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
*/
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = fueword32(&cv->c_has_waiters, &hasw);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == 0 && hasw == 0)
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
suword32(&cv->c_has_waiters, 1);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&uq->uq_key);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
error = do_unlock_umutex(td, m, false);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
Remove trailing semicolon, it is a typo.
2012-03-30 12:57:14 +00:00
if (timeout != NULL)
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
abs_timeout_init(&timo, clockid, (wflags & CVWAIT_ABSTIME) != 0,
timeout);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
umtxq_lock(&uq->uq_key);
if (error == 0) {
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = umtxq_sleep(uq, "ucond", timeout == NULL ?
NULL : &timo);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
}
If a thread is removed from umtxq while sleeping, reset error code to zero, this gives userland a better indication that a thread needn't to be cancelled.
2010-08-25 03:14:32 +00:00
if ((uq->uq_flags & UQF_UMTXQ) == 0)
error = 0;
else {
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
/*
* This must be timeout,interrupted by signal or
* surprious wakeup, clear c_has_waiter flag when
* necessary.
*/
umtxq_busy(&uq->uq_key);
if ((uq->uq_flags & UQF_UMTXQ) != 0) {
int oldlen = uq->uq_cur_queue->length;
umtxq_remove(uq);
if (oldlen == 1) {
umtxq_unlock(&uq->uq_key);
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
suword32(&cv->c_has_waiters, 0);
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
umtxq_lock(&uq->uq_key);
}
}
umtxq_unbusy(&uq->uq_key);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
if (error == ERESTART)
error = EINTR;
}
If a thread is removed from umtxq while sleeping, reset error code to zero, this gives userland a better indication that a thread needn't to be cancelled.
2010-08-25 03:14:32 +00:00
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
return (error);
}
/*
* Signal a userland condition variable.
*/
static int
do_cv_signal(struct thread *td, struct ucond *cv)
{
struct umtx_key key;
int error, cnt, nwake;
uint32_t flags;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&cv->c_flags, &flags);
if (error == -1)
return (EFAULT);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0)
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
cnt = umtxq_count(&key);
nwake = umtxq_signal(&key, 1);
if (cnt <= nwake) {
umtxq_unlock(&key);
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = suword32(&cv->c_has_waiters, 0);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
error = EFAULT;
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
umtxq_lock(&key);
}
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
return (error);
}
static int
do_cv_broadcast(struct thread *td, struct ucond *cv)
{
struct umtx_key key;
int error;
uint32_t flags;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&cv->c_flags, &flags);
if (error == -1)
return (EFAULT);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0)
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
umtxq_signal(&key, INT_MAX);
umtxq_unlock(&key);
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = suword32(&cv->c_has_waiters, 0);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
error = EFAULT;
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&key);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
umtx_key_release(&key);
return (error);
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
static int
Style. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2019-06-28 20:40:54 +00:00
do_rw_rdlock(struct thread *td, struct urwlock *rwlock, long fflag,
struct _umtx_time *timeout)
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
{
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout timo;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
struct umtx_q *uq;
uint32_t flags, wrflags;
int32_t state, oldstate;
int32_t blocked_readers;
In both do_rw_wrlock() and do_rw_rdlock() after r304808, do not obliterate possible error from sleep with errors from umtxq_check_susp(), when looping to clear URWLOCK_{READ,WRITE}_WAITERS. Noted and reviewed by: vangyzen Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 19:15:02 +00:00
int error, error1, rv;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
uq = td->td_umtxq;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&rwlock->rw_flags, &flags);
if (error == -1)
return (EFAULT);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
if (error != 0)
return (error);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
if (timeout != NULL)
abs_timeout_init2(&timo, timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
wrflags = URWLOCK_WRITE_OWNER;
if (!(fflag & URWLOCK_PREFER_READER) && !(flags & URWLOCK_PREFER_READER))
wrflags |= URWLOCK_WRITE_WAITERS;
for (;;) {
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&rwlock->rw_state, &state);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1) {
umtx_key_release(&uq->uq_key);
return (EFAULT);
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
/* try to lock it */
while (!(state & wrflags)) {
More style. Re-wrap long lines, reformat comments, remove excessive blank line. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2019-07-02 21:03:06 +00:00
if (__predict_false(URWLOCK_READER_COUNT(state) ==
URWLOCK_MAX_READERS)) {
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtx_key_release(&uq->uq_key);
return (EAGAIN);
}
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&rwlock->rw_state, state,
&oldstate, state + 1);
if (rv == -1) {
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
umtx_key_release(&uq->uq_key);
return (EFAULT);
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(oldstate == state);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtx_key_release(&uq->uq_key);
return (0);
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, true);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
break;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
state = oldstate;
}
if (error)
break;
/* grab monitor lock */
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
After busied the lock, re-read state word before checking waiters flag, otherwise, the waiters bit may not be set and a wakeup is lost. Submitted by: justin.teller at gmail dot com MFC after: 3 days
2010-02-03 03:56:32 +00:00
/*
* re-read the state, in case it changed between the try-lock above
* and the check below
*/
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&rwlock->rw_state, &state);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1)
error = EFAULT;
After busied the lock, re-read state word before checking waiters flag, otherwise, the waiters bit may not be set and a wakeup is lost. Submitted by: justin.teller at gmail dot com MFC after: 3 days
2010-02-03 03:56:32 +00:00
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
/* set read contention bit */
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
while (error == 0 && (state & wrflags) &&
!(state & URWLOCK_READ_WAITERS)) {
rv = casueword32(&rwlock->rw_state, state,
&oldstate, state | URWLOCK_READ_WAITERS);
if (rv == -1) {
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
error = EFAULT;
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(oldstate == state);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
goto sleep;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
state = oldstate;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, false);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
break;
}
if (error != 0) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&uq->uq_key);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
break;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
/* state is changed while setting flags, restart */
if (!(state & wrflags)) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&uq->uq_key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, true);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
break;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
continue;
}
sleep:
More style. Re-wrap long lines, reformat comments, remove excessive blank line. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2019-07-02 21:03:06 +00:00
/*
* Contention bit is set, before sleeping, increase
* read waiter count.
*/
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = fueword32(&rwlock->rw_blocked_readers,
&blocked_readers);
if (rv == -1) {
umtxq_unbusy_unlocked(&uq->uq_key);
error = EFAULT;
break;
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
suword32(&rwlock->rw_blocked_readers, blocked_readers+1);
while (state & wrflags) {
umtxq_lock(&uq->uq_key);
umtxq_insert(uq);
umtxq_unbusy(&uq->uq_key);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = umtxq_sleep(uq, "urdlck", timeout == NULL ?
NULL : &timo);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_busy(&uq->uq_key);
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
if (error)
break;
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&rwlock->rw_state, &state);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1) {
error = EFAULT;
break;
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
/* decrease read waiter count, and may clear read contention bit */
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = fueword32(&rwlock->rw_blocked_readers,
&blocked_readers);
if (rv == -1) {
umtxq_unbusy_unlocked(&uq->uq_key);
error = EFAULT;
break;
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
suword32(&rwlock->rw_blocked_readers, blocked_readers-1);
if (blocked_readers == 1) {
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&rwlock->rw_state, &state);
Prevent leak of URWLOCK_READ_WAITERS flag for urwlocks. If there was some error, e.g. the sleep was interrupted, as in the referenced PR, do_rw_rdlock() did not cleared URWLOCK_READ_WAITERS. Since unlock only wakes up write waiters when there is no read waiters, for URWLOCK_PREFER_READER kind of locks, the result was missed wakeups for writers. In particular, the most visible victims are ld-elf.so locks in processes which loaded libthr, because rtld locks are urwlocks in prefer-reader mode. Normal rwlocks fall into prefer-reader mode only if thread already owns rw lock in read mode, which is not typical and correspondingly less visible. In the PR, unowned rtld bind lock was waited for in the process where only one thread was left alive. Note that do_rw_wrlock() correctly clears URWLOCK_WRITE_WAITERS in case of errors. Reported and tested by: longwitz@incore.de PR: 211947 Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 16:35:42 +00:00
if (rv == -1) {
umtxq_unbusy_unlocked(&uq->uq_key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = EFAULT;
Prevent leak of URWLOCK_READ_WAITERS flag for urwlocks. If there was some error, e.g. the sleep was interrupted, as in the referenced PR, do_rw_rdlock() did not cleared URWLOCK_READ_WAITERS. Since unlock only wakes up write waiters when there is no read waiters, for URWLOCK_PREFER_READER kind of locks, the result was missed wakeups for writers. In particular, the most visible victims are ld-elf.so locks in processes which loaded libthr, because rtld locks are urwlocks in prefer-reader mode. Normal rwlocks fall into prefer-reader mode only if thread already owns rw lock in read mode, which is not typical and correspondingly less visible. In the PR, unowned rtld bind lock was waited for in the process where only one thread was left alive. Note that do_rw_wrlock() correctly clears URWLOCK_WRITE_WAITERS in case of errors. Reported and tested by: longwitz@incore.de PR: 211947 Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 16:35:42 +00:00
break;
}
for (;;) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&rwlock->rw_state, state,
&oldstate, state & ~URWLOCK_READ_WAITERS);
if (rv == -1) {
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
error = EFAULT;
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(oldstate == state);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
break;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
state = oldstate;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error1 = umtxq_check_susp(td, false);
In both do_rw_wrlock() and do_rw_rdlock() after r304808, do not obliterate possible error from sleep with errors from umtxq_check_susp(), when looping to clear URWLOCK_{READ,WRITE}_WAITERS. Noted and reviewed by: vangyzen Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 19:15:02 +00:00
if (error1 != 0) {
if (error == 0)
error = error1;
Prevent leak of URWLOCK_READ_WAITERS flag for urwlocks. If there was some error, e.g. the sleep was interrupted, as in the referenced PR, do_rw_rdlock() did not cleared URWLOCK_READ_WAITERS. Since unlock only wakes up write waiters when there is no read waiters, for URWLOCK_PREFER_READER kind of locks, the result was missed wakeups for writers. In particular, the most visible victims are ld-elf.so locks in processes which loaded libthr, because rtld locks are urwlocks in prefer-reader mode. Normal rwlocks fall into prefer-reader mode only if thread already owns rw lock in read mode, which is not typical and correspondingly less visible. In the PR, unowned rtld bind lock was waited for in the process where only one thread was left alive. Note that do_rw_wrlock() correctly clears URWLOCK_WRITE_WAITERS in case of errors. Reported and tested by: longwitz@incore.de PR: 211947 Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 16:35:42 +00:00
break;
In both do_rw_wrlock() and do_rw_rdlock() after r304808, do not obliterate possible error from sleep with errors from umtxq_check_susp(), when looping to clear URWLOCK_{READ,WRITE}_WAITERS. Noted and reviewed by: vangyzen Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 19:15:02 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
}
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&uq->uq_key);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
break;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
umtx_key_release(&uq->uq_key);
Er, don't restart a timeout version.
2008-04-02 04:26:59 +00:00
if (error == ERESTART)
error = EINTR;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
return (error);
}
static int
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
do_rw_wrlock(struct thread *td, struct urwlock *rwlock, struct _umtx_time *timeout)
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
{
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout timo;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
struct umtx_q *uq;
uint32_t flags;
int32_t state, oldstate;
int32_t blocked_writers;
In function do_rw_wrlock, when a writer got an error and before returning, check if there are readers blocked by us via URWLOCK_WRITE_WAITERS flag, and resume the readers. The error must be EAGAIN, otherwise there must have memory problem, and nobody can rescue the buggy application. The revision 197445 might be reverted.
2009-09-25 00:03:13 +00:00
int32_t blocked_readers;
In both do_rw_wrlock() and do_rw_rdlock() after r304808, do not obliterate possible error from sleep with errors from umtxq_check_susp(), when looping to clear URWLOCK_{READ,WRITE}_WAITERS. Noted and reviewed by: vangyzen Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 19:15:02 +00:00
int error, error1, rv;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
uq = td->td_umtxq;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&rwlock->rw_flags, &flags);
if (error == -1)
return (EFAULT);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
if (error != 0)
return (error);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
if (timeout != NULL)
abs_timeout_init2(&timo, timeout);
In function do_rw_wrlock, when a writer got an error and before returning, check if there are readers blocked by us via URWLOCK_WRITE_WAITERS flag, and resume the readers. The error must be EAGAIN, otherwise there must have memory problem, and nobody can rescue the buggy application. The revision 197445 might be reverted.
2009-09-25 00:03:13 +00:00
blocked_readers = 0;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
for (;;) {
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&rwlock->rw_state, &state);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1) {
umtx_key_release(&uq->uq_key);
return (EFAULT);
}
More style. Re-wrap long lines, reformat comments, remove excessive blank line. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2019-07-02 21:03:06 +00:00
while ((state & URWLOCK_WRITE_OWNER) == 0 &&
URWLOCK_READER_COUNT(state) == 0) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&rwlock->rw_state, state,
&oldstate, state | URWLOCK_WRITE_OWNER);
if (rv == -1) {
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
umtx_key_release(&uq->uq_key);
return (EFAULT);
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(oldstate == state);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtx_key_release(&uq->uq_key);
return (0);
}
state = oldstate;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, true);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
break;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
In function do_rw_wrlock, when a writer got an error and before returning, check if there are readers blocked by us via URWLOCK_WRITE_WAITERS flag, and resume the readers. The error must be EAGAIN, otherwise there must have memory problem, and nobody can rescue the buggy application. The revision 197445 might be reverted.
2009-09-25 00:03:13 +00:00
if (error) {
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if ((state & (URWLOCK_WRITE_OWNER |
URWLOCK_WRITE_WAITERS)) == 0 &&
In function do_rw_wrlock, when a writer got an error and before returning, check if there are readers blocked by us via URWLOCK_WRITE_WAITERS flag, and resume the readers. The error must be EAGAIN, otherwise there must have memory problem, and nobody can rescue the buggy application. The revision 197445 might be reverted.
2009-09-25 00:03:13 +00:00
blocked_readers != 0) {
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
umtxq_signal_queue(&uq->uq_key, INT_MAX,
UMTX_SHARED_QUEUE);
In function do_rw_wrlock, when a writer got an error and before returning, check if there are readers blocked by us via URWLOCK_WRITE_WAITERS flag, and resume the readers. The error must be EAGAIN, otherwise there must have memory problem, and nobody can rescue the buggy application. The revision 197445 might be reverted.
2009-09-25 00:03:13 +00:00
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
break;
In function do_rw_wrlock, when a writer got an error and before returning, check if there are readers blocked by us via URWLOCK_WRITE_WAITERS flag, and resume the readers. The error must be EAGAIN, otherwise there must have memory problem, and nobody can rescue the buggy application. The revision 197445 might be reverted.
2009-09-25 00:03:13 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
/* grab monitor lock */
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
After busied the lock, re-read state word before checking waiters flag, otherwise, the waiters bit may not be set and a wakeup is lost. Submitted by: justin.teller at gmail dot com MFC after: 3 days
2010-02-03 03:56:32 +00:00
/*
More style. Re-wrap long lines, reformat comments, remove excessive blank line. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2019-07-02 21:03:06 +00:00
* Re-read the state, in case it changed between the
* try-lock above and the check below.
After busied the lock, re-read state word before checking waiters flag, otherwise, the waiters bit may not be set and a wakeup is lost. Submitted by: justin.teller at gmail dot com MFC after: 3 days
2010-02-03 03:56:32 +00:00
*/
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&rwlock->rw_state, &state);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1)
error = EFAULT;
After busied the lock, re-read state word before checking waiters flag, otherwise, the waiters bit may not be set and a wakeup is lost. Submitted by: justin.teller at gmail dot com MFC after: 3 days
2010-02-03 03:56:32 +00:00
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
while (error == 0 && ((state & URWLOCK_WRITE_OWNER) ||
URWLOCK_READER_COUNT(state) != 0) &&
(state & URWLOCK_WRITE_WAITERS) == 0) {
rv = casueword32(&rwlock->rw_state, state,
&oldstate, state | URWLOCK_WRITE_WAITERS);
if (rv == -1) {
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
error = EFAULT;
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(oldstate == state);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
goto sleep;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
state = oldstate;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, false);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
break;
}
if (error != 0) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&uq->uq_key);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
break;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
Style. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2019-07-02 19:32:48 +00:00
if ((state & URWLOCK_WRITE_OWNER) == 0 &&
URWLOCK_READER_COUNT(state) == 0) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&uq->uq_key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, false);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
break;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
continue;
}
sleep:
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = fueword32(&rwlock->rw_blocked_writers,
&blocked_writers);
if (rv == -1) {
umtxq_unbusy_unlocked(&uq->uq_key);
error = EFAULT;
break;
}
Style. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2019-07-02 19:32:48 +00:00
suword32(&rwlock->rw_blocked_writers, blocked_writers + 1);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
Style. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2019-07-02 19:32:48 +00:00
while ((state & URWLOCK_WRITE_OWNER) ||
URWLOCK_READER_COUNT(state) != 0) {
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_lock(&uq->uq_key);
umtxq_insert_queue(uq, UMTX_EXCLUSIVE_QUEUE);
umtxq_unbusy(&uq->uq_key);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = umtxq_sleep(uq, "uwrlck", timeout == NULL ?
NULL : &timo);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
umtxq_busy(&uq->uq_key);
umtxq_remove_queue(uq, UMTX_EXCLUSIVE_QUEUE);
umtxq_unlock(&uq->uq_key);
if (error)
break;
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&rwlock->rw_state, &state);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1) {
error = EFAULT;
break;
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = fueword32(&rwlock->rw_blocked_writers,
&blocked_writers);
if (rv == -1) {
umtxq_unbusy_unlocked(&uq->uq_key);
error = EFAULT;
break;
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
suword32(&rwlock->rw_blocked_writers, blocked_writers-1);
if (blocked_writers == 1) {
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&rwlock->rw_state, &state);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1) {
umtxq_unbusy_unlocked(&uq->uq_key);
error = EFAULT;
break;
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
for (;;) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&rwlock->rw_state, state,
&oldstate, state & ~URWLOCK_WRITE_WAITERS);
if (rv == -1) {
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
error = EFAULT;
break;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0) {
MPASS(oldstate == state);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
break;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
state = oldstate;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error1 = umtxq_check_susp(td, false);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
/*
* We are leaving the URWLOCK_WRITE_WAITERS
* behind, but this should not harm the
* correctness.
*/
In both do_rw_wrlock() and do_rw_rdlock() after r304808, do not obliterate possible error from sleep with errors from umtxq_check_susp(), when looping to clear URWLOCK_{READ,WRITE}_WAITERS. Noted and reviewed by: vangyzen Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 19:15:02 +00:00
if (error1 != 0) {
if (error == 0)
error = error1;
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
break;
In both do_rw_wrlock() and do_rw_rdlock() after r304808, do not obliterate possible error from sleep with errors from umtxq_check_susp(), when looping to clear URWLOCK_{READ,WRITE}_WAITERS. Noted and reviewed by: vangyzen Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-08-25 19:15:02 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = fueword32(&rwlock->rw_blocked_readers,
&blocked_readers);
if (rv == -1) {
umtxq_unbusy_unlocked(&uq->uq_key);
error = EFAULT;
break;
}
In function do_rw_wrlock, when a writer got an error and before returning, check if there are readers blocked by us via URWLOCK_WRITE_WAITERS flag, and resume the readers. The error must be EAGAIN, otherwise there must have memory problem, and nobody can rescue the buggy application. The revision 197445 might be reverted.
2009-09-25 00:03:13 +00:00
} else
blocked_readers = 0;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
umtxq_unbusy_unlocked(&uq->uq_key);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
umtx_key_release(&uq->uq_key);
Er, don't restart a timeout version.
2008-04-02 04:26:59 +00:00
if (error == ERESTART)
error = EINTR;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
return (error);
}
static int
let umtxq_busy() only spin on mp machine. make function name do_rwlock_unlock to be consistent with others.
2008-04-03 11:49:20 +00:00
do_rw_unlock(struct thread *td, struct urwlock *rwlock)
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
{
struct umtx_q *uq;
uint32_t flags;
int32_t state, oldstate;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
int error, rv, q, count;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
uq = td->td_umtxq;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&rwlock->rw_flags, &flags);
if (error == -1)
return (EFAULT);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
if (error != 0)
return (error);
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = fueword32(&rwlock->rw_state, &state);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1) {
error = EFAULT;
goto out;
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
if (state & URWLOCK_WRITE_OWNER) {
for (;;) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&rwlock->rw_state, state,
&oldstate, state & ~URWLOCK_WRITE_OWNER);
if (rv == -1) {
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
error = EFAULT;
goto out;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 1) {
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
state = oldstate;
if (!(oldstate & URWLOCK_WRITE_OWNER)) {
error = EPERM;
goto out;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, true);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
goto out;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
} else
break;
}
} else if (URWLOCK_READER_COUNT(state) != 0) {
for (;;) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&rwlock->rw_state, state,
&oldstate, state - 1);
if (rv == -1) {
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
error = EFAULT;
goto out;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 1) {
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
state = oldstate;
if (URWLOCK_READER_COUNT(oldstate) == 0) {
error = EPERM;
goto out;
}
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
error = umtxq_check_susp(td, true);
Fix two issues with the spin loops in the umtx(2) implementation. - When looping, check for the pending suspension. Otherwise, other usermode thread which races with the looping one, could try to prevent the process from stopping or exiting. - Add missed checks for the faults from casuword*(). The code is structured in a way which makes the loops exit if the specified address is invalid, since both fuword() and casuword() return -1 on the fault. But if the address is mapped readonly, the typical value read by fuword() is different from -1, while casuword() returns -1. Absent the checks for casuword() faults, this is interpreted as the race with other thread and causes non-interruptible spinning in the kernel. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-06-13 09:33:22 +00:00
if (error != 0)
goto out;
} else
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
break;
}
} else {
error = EPERM;
goto out;
}
count = 0;
if (!(flags & URWLOCK_PREFER_READER)) {
if (state & URWLOCK_WRITE_WAITERS) {
count = 1;
q = UMTX_EXCLUSIVE_QUEUE;
} else if (state & URWLOCK_READ_WAITERS) {
count = INT_MAX;
q = UMTX_SHARED_QUEUE;
}
} else {
if (state & URWLOCK_READ_WAITERS) {
count = INT_MAX;
q = UMTX_SHARED_QUEUE;
} else if (state & URWLOCK_WRITE_WAITERS) {
count = 1;
q = UMTX_EXCLUSIVE_QUEUE;
}
}
if (count) {
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_signal_queue(&uq->uq_key, count, q);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
}
out:
umtx_key_release(&uq->uq_key);
return (error);
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#if defined(COMPAT_FREEBSD9) || defined(COMPAT_FREEBSD10)
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
static int
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
do_sem_wait(struct thread *td, struct _usem *sem, struct _umtx_time *timeout)
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
{
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
struct abs_timeout timo;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
struct umtx_q *uq;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
uint32_t flags, count, count1;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
int error, rv, rv1;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
uq = td->td_umtxq;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&sem->_flags, &flags);
if (error == -1)
return (EFAULT);
Add key type TYPE_SEM.
2010-01-09 06:05:31 +00:00
error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &uq->uq_key);
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
if (error != 0)
return (error);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
if (timeout != NULL)
abs_timeout_init2(&timo, timeout);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
again:
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_insert(uq);
umtxq_unlock(&uq->uq_key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&sem->_has_waiters, 0, &count1, 1);
if (rv == 0)
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
rv1 = fueword32(&sem->_count, &count);
Fix handling of transient casueword(9) failures in do_sem_wait(). In particular, restart should be only done when the failure is transient. For this, recheck the count1 value after the operation. Note that do_sem_wait() is older usem interface. Reported and tested by: bdrewery Sponsored by: The FreeBSD Foundation MFC after: 1 week
2019-07-31 19:16:49 +00:00
if (rv == -1 || (rv == 0 && (rv1 == -1 || count != 0)) ||
(rv == 1 && count1 == 0)) {
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 1) {
rv = umtxq_check_susp(td, true);
if (rv == 0)
goto again;
error = rv;
goto out;
}
if (rv == 0)
rv = rv1;
error = rv == -1 ? EFAULT : 0;
goto out;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
}
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = umtxq_sleep(uq, "usem", timeout == NULL ? NULL : &timo);
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
If a thread is removed from umtxq while sleeping, reset error code to zero, this gives userland a better indication that a thread needn't to be cancelled.
2010-08-25 03:14:32 +00:00
if ((uq->uq_flags & UQF_UMTXQ) == 0)
error = 0;
else {
umtxq_remove(uq);
sem: Restart the POSIX sem_* calls after signals with SA_RESTART set. Programs often do not expect an [EINTR] return from sem_wait() and POSIX only allows it if the signal was installed without SA_RESTART. The timeout in sem_timedwait() is absolute so it can be restarted normally. The umtx call can be invoked with a relative timeout and in that case [ERESTART] must be changed to [EINTR]. However, libc does not do this. The old POSIX semaphore implementation did this correctly (before r249566), unlike the new umtx one. It may be desirable to avoid [EINTR] completely, which matches the pthread functions and is explicitly permitted by POSIX. However, the kernel must return [EINTR] at least for signals with SA_RESTART clear, otherwise pthread cancellation will not abort a semaphore wait. In this commit, only restore the 8.x behaviour which is also permitted by POSIX. Discussed with: jhb MFC after: 1 week
2013-04-19 10:16:00 +00:00
/* A relative timeout cannot be restarted. */
if (error == ERESTART && timeout != NULL &&
(timeout->_flags & UMTX_ABSTIME) == 0)
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
error = EINTR;
}
umtxq_unlock(&uq->uq_key);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
out:
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
umtx_key_release(&uq->uq_key);
return (error);
}
/*
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
* Signal a userland semaphore.
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
*/
static int
do_sem_wake(struct thread *td, struct _usem *sem)
{
struct umtx_key key;
In sem_post, the field _has_waiters is no longer used, because some application destroys semaphore after sem_wait returns. Just enter kernel to wake up sleeping threads, only update _has_waiters if it is safe. While here, check if the value exceed SEM_VALUE_MAX and return EOVERFLOW if this is true.
2012-04-05 03:05:02 +00:00
int error, cnt;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
uint32_t flags;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
error = fueword32(&sem->_flags, &flags);
if (error == -1)
return (EFAULT);
Add key type TYPE_SEM.
2010-01-09 06:05:31 +00:00
if ((error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &key)) != 0)
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
cnt = umtxq_count(&key);
In sem_post, the field _has_waiters is no longer used, because some application destroys semaphore after sem_wait returns. Just enter kernel to wake up sleeping threads, only update _has_waiters if it is safe. While here, check if the value exceed SEM_VALUE_MAX and return EOVERFLOW if this is true.
2012-04-05 03:05:02 +00:00
if (cnt > 0) {
/*
* Check if count is greater than 0, this means the memory is
* still being referenced by user code, so we can safely
* update _has_waiters flag.
*/
if (cnt == 1) {
umtxq_unlock(&key);
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
error = suword32(&sem->_has_waiters, 0);
In sem_post, the field _has_waiters is no longer used, because some application destroys semaphore after sem_wait returns. Just enter kernel to wake up sleeping threads, only update _has_waiters if it is safe. While here, check if the value exceed SEM_VALUE_MAX and return EOVERFLOW if this is true.
2012-04-05 03:05:02 +00:00
umtxq_lock(&key);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (error == -1)
error = EFAULT;
In sem_post, the field _has_waiters is no longer used, because some application destroys semaphore after sem_wait returns. Just enter kernel to wake up sleeping threads, only update _has_waiters if it is safe. While here, check if the value exceed SEM_VALUE_MAX and return EOVERFLOW if this is true.
2012-04-05 03:05:02 +00:00
}
sem_post(): wake up the sleeper only after adjusting has_waiters If the caller of sem_post() wakes up a thread sleeping via sem_wait() before it clears the has_waiters flag, the caller of sem_wait() has no way of knowing when it is safe to destroy the semaphore and reuse the memory. This is because the caller of sem_post() may be interrupted between the wake step and the clearing of has_waiters. It will then write into the has_waiters flag in userspace after being preempted for some unknown amount of time. Reviewed by: jhb, kib, vangyzen Approved by: kib (mentor), vangyzen (mentor) MFC after: 2 weeks Sponsored by: Dell Inc. Differential Revision: https://reviews.freebsd.org/D7505
2016-08-15 20:09:09 +00:00
umtxq_signal(&key, 1);
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
}
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
return (error);
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#endif
static int
do_sem2_wait(struct thread *td, struct _usem2 *sem, struct _umtx_time *timeout)
{
struct abs_timeout timo;
struct umtx_q *uq;
uint32_t count, flags;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
int error, rv;
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
uq = td->td_umtxq;
flags = fuword32(&sem->_flags);
if (timeout != NULL)
abs_timeout_init2(&timo, timeout);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
again:
In do_sem2_wait(), balance umtx_key_get() with umtx_key_release() on retry. Reported by: ler Bisected and reviewed by: markj Sponsored by: The FreeBSD Foundation MFC after: 12 days
2019-07-15 19:18:25 +00:00
error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &uq->uq_key);
if (error != 0)
return (error);
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_insert(uq);
umtxq_unlock(&uq->uq_key);
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&sem->_count, &count);
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1) {
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
return (EFAULT);
}
for (;;) {
if (USEM_COUNT(count) != 0) {
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
return (0);
}
if (count == USEM_HAS_WAITERS)
break;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&sem->_count, 0, &count, USEM_HAS_WAITERS);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 0)
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
break;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
if (rv == -1)
return (EFAULT);
rv = umtxq_check_susp(td, true);
if (rv != 0)
return (rv);
goto again;
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
}
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
error = umtxq_sleep(uq, "usem", timeout == NULL ? NULL : &timo);
if ((uq->uq_flags & UQF_UMTXQ) == 0)
error = 0;
else {
umtxq_remove(uq);
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
if (timeout != NULL && (timeout->_flags & UMTX_ABSTIME) == 0) {
/* A relative timeout cannot be restarted. */
if (error == ERESTART)
error = EINTR;
if (error == EINTR) {
abs_timeout_update(&timo);
Make timespecadd(3) and friends public The timespecadd(3) family of macros were imported from NetBSD back in r35029. However, they were initially guarded by #ifdef _KERNEL. In the meantime, we have grown at least 28 syscalls that use timespecs in some way, leading many programs both inside and outside of the base system to redefine those macros. It's better just to make the definitions public. Our kernel currently defines two-argument versions of timespecadd and timespecsub. NetBSD, OpenBSD, and FreeDesktop.org's libbsd, however, define three-argument versions. Solaris also defines a three-argument version, but only in its kernel. This revision changes our definition to match the common three-argument version. Bump _FreeBSD_version due to the breaking KPI change. Discussed with: cem, jilles, ian, bde Differential Revision: https://reviews.freebsd.org/D14725
2018-07-30 15:46:40 +00:00
timespecsub(&timo.end, &timo.cur,
&timeout->_timeout);
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
}
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
}
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
return (error);
}
/*
* Signal a userland semaphore.
*/
static int
do_sem2_wake(struct thread *td, struct _usem2 *sem)
{
struct umtx_key key;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
int error, cnt, rv;
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
uint32_t count, flags;
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = fueword32(&sem->_flags, &flags);
if (rv == -1)
return (EFAULT);
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
if ((error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &key)) != 0)
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
cnt = umtxq_count(&key);
if (cnt > 0) {
/*
* If this was the last sleeping thread, clear the waiters
* flag in _count.
*/
if (cnt == 1) {
umtxq_unlock(&key);
Add type qualifier volatile to the base (userspace) address argument of fuword(9) and suword(9). This makes the functions type-compatible with volatile objects and does not require devolatile force, e.g. in kern_umtx.c. Requested by: bde Reviewed by: jhb Sponsored by: The FreeBSD Foundation MFC after: 3 weeks
2014-10-31 17:43:21 +00:00
rv = fueword32(&sem->_count, &count);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
while (rv != -1 && count & USEM_HAS_WAITERS) {
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
rv = casueword32(&sem->_count, count, &count,
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
count & ~USEM_HAS_WAITERS);
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
if (rv == 1) {
rv = umtxq_check_susp(td, true);
if (rv != 0)
break;
}
}
Convert kern_umtx.c to use fueword() and casueword(). Also fix some mishandling of suword(9) errors as errno, which resulted in spurious ERESTART. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 3 weeks
2014-10-28 15:30:33 +00:00
if (rv == -1)
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
error = EFAULT;
Provide protection against starvation of the ll/sc loops when accessing userpace. Casueword(9) on ll/sc architectures must be prepared for userspace constantly modifying the same cache line as containing the CAS word, and not loop infinitely. Otherwise, rogue userspace livelocks the kernel. To fix the issue, change casueword(9) interface to return new value 1 indicating that either comparision or store failed, instead of relying on the oldval == *oldvalp comparison. The primitive no longer retries the operation if it failed spuriously. Modify callers of casueword(9), all in kern_umtx.c, to handle retries, and react to stops and requests to terminate between retries. On x86, despite cmpxchg should not return spurious failures, we can take advantage of the new interface and just return PSL.ZF. Reviewed by: andrew (arm64, previous version), markj Tested by: pho Reported by: https://xenbits.xen.org/xsa/advisory-295.txt Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D20772
2019-07-12 18:43:24 +00:00
else if (rv > 0) {
error = rv;
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
umtxq_lock(&key);
}
sem_post(): wake up the sleeper only after adjusting has_waiters If the caller of sem_post() wakes up a thread sleeping via sem_wait() before it clears the has_waiters flag, the caller of sem_wait() has no way of knowing when it is safe to destroy the semaphore and reuse the memory. This is because the caller of sem_post() may be interrupted between the wake step and the clearing of has_waiters. It will then write into the has_waiters flag in userspace after being preempted for some unknown amount of time. Reviewed by: jhb, kib, vangyzen Approved by: kib (mentor), vangyzen (mentor) MFC after: 2 weeks Sponsored by: Dell Inc. Differential Revision: https://reviews.freebsd.org/D7505
2016-08-15 20:09:09 +00:00
umtxq_signal(&key, 1);
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
}
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
return (error);
}
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
Add umtx_copyin_timeout() and move parameter checks here. In collaboration with: kib MFC after: 1 week
2011-12-03 12:30:58 +00:00
inline int
umtx_copyin_timeout(const void *addr, struct timespec *tsp)
{
int error;
error = copyin(addr, tsp, sizeof(struct timespec));
if (error == 0) {
if (tsp->tv_sec < 0 ||
tsp->tv_nsec >= 1000000000 ||
tsp->tv_nsec < 0)
error = EINVAL;
}
return (error);
}
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
static inline int
umtx_copyin_umtx_time(const void *addr, size_t size, struct _umtx_time *tp)
{
int error;
initialize clock ID and flags only when copying timespec, a _umtx_time copy already contains these fields.
2012-02-29 02:01:48 +00:00
if (size <= sizeof(struct timespec)) {
tp->_clockid = CLOCK_REALTIME;
tp->_flags = 0;
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = copyin(addr, &tp->_timeout, sizeof(struct timespec));
initialize clock ID and flags only when copying timespec, a _umtx_time copy already contains these fields.
2012-02-29 02:01:48 +00:00
} else
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = copyin(addr, tp, sizeof(struct _umtx_time));
if (error != 0)
return (error);
if (tp->_timeout.tv_sec < 0 ||
tp->_timeout.tv_nsec >= 1000000000 || tp->_timeout.tv_nsec < 0)
return (EINVAL);
return (0);
}
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
static int
Remove dead code from umtx support: - Retire long time unused (basically always unused) sys__umtx_lock() and sys__umtx_unlock() syscalls - struct umtx and their supporting definitions - UMUTEX_ERROR_CHECK flag - Retire UMTX_OP_LOCK/UMTX_OP_UNLOCK from _umtx_op() syscall __FreeBSD_version is not bumped yet because it is expected that further breakages to the umtx interface will follow up in the next days. However there will be a final bump when necessary. Sponsored by: EMC / Isilon storage division Reviewed by: jhb
2014-03-18 21:32:03 +00:00
__umtx_op_unimpl(struct thread *td, struct _umtx_op_args *uap)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
Remove dead code from umtx support: - Retire long time unused (basically always unused) sys__umtx_lock() and sys__umtx_unlock() syscalls - struct umtx and their supporting definitions - UMUTEX_ERROR_CHECK flag - Retire UMTX_OP_LOCK/UMTX_OP_UNLOCK from _umtx_op() syscall __FreeBSD_version is not bumped yet because it is expected that further breakages to the umtx interface will follow up in the next days. However there will be a final bump when necessary. Sponsored by: EMC / Isilon storage division Reviewed by: jhb
2014-03-18 21:32:03 +00:00
return (EOPNOTSUPP);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
static int
__umtx_op_wait(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time timeout, *tm_p;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
int error;
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time(
uap->uaddr2, (size_t)uap->uaddr1, &timeout);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_wait(td, uap->obj, uap->val, tm_p, 0, 0));
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
Add function UMTX_OP_WAIT_UINT, the function causes thread to wait for an integer to be changed.
2007-11-21 04:21:02 +00:00
static int
__umtx_op_wait_uint(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time timeout, *tm_p;
Add function UMTX_OP_WAIT_UINT, the function causes thread to wait for an integer to be changed.
2007-11-21 04:21:02 +00:00
int error;
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add function UMTX_OP_WAIT_UINT, the function causes thread to wait for an integer to be changed.
2007-11-21 04:21:02 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time(
uap->uaddr2, (size_t)uap->uaddr1, &timeout);
Add function UMTX_OP_WAIT_UINT, the function causes thread to wait for an integer to be changed.
2007-11-21 04:21:02 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add function UMTX_OP_WAIT_UINT, the function causes thread to wait for an integer to be changed.
2007-11-21 04:21:02 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_wait(td, uap->obj, uap->val, tm_p, 1, 0));
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
}
static int
__umtx_op_wait_uint_private(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
int error;
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time(
uap->uaddr2, (size_t)uap->uaddr1, &timeout);
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_wait(td, uap->obj, uap->val, tm_p, 1, 1));
Add function UMTX_OP_WAIT_UINT, the function causes thread to wait for an integer to be changed.
2007-11-21 04:21:02 +00:00
}
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
static int
__umtx_op_wake(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
return (kern_umtx_wake(td, uap->obj, uap->val, 0));
}
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
#define BATCH_SIZE 128
static int
__umtx_op_nwake_private(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
char *uaddrs[BATCH_SIZE], **upp;
int count, error, i, pos, tocopy;
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
upp = (char **)uap->obj;
error = 0;
for (count = uap->val, pos = 0; count > 0; count -= tocopy,
pos += tocopy) {
tocopy = MIN(count, BATCH_SIZE);
error = copyin(upp + pos, uaddrs, tocopy * sizeof(char *));
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
if (error != 0)
break;
for (i = 0; i < tocopy; ++i)
kern_umtx_wake(td, uaddrs[i], INT_MAX, 1);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
maybe_yield();
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
}
return (error);
}
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
static int
__umtx_op_wake_private(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
return (kern_umtx_wake(td, uap->obj, uap->val, 1));
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
static int
__umtx_op_lock_umutex(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time(
uap->uaddr2, (size_t)uap->uaddr1, &timeout);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_lock_umutex(td, uap->obj, tm_p, 0));
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
static int
__umtx_op_trylock_umutex(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_lock_umutex(td, uap->obj, NULL, _UMUTEX_TRY));
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
}
static int
__umtx_op_wait_umutex(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time(
uap->uaddr2, (size_t)uap->uaddr1, &timeout);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_lock_umutex(td, uap->obj, tm_p, _UMUTEX_WAIT));
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
}
static int
__umtx_op_wake_umutex(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_wake_umutex(td, uap->obj));
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
static int
__umtx_op_unlock_umutex(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_unlock_umutex(td, uap->obj, false));
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
static int
__umtx_op_set_ceiling(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_set_ceiling(td, uap->obj, uap->val, uap->uaddr1));
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
static int
__umtx_op_cv_wait(struct thread *td, struct _umtx_op_args *uap)
{
struct timespec *ts, timeout;
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL)
ts = NULL;
else {
Add umtx_copyin_timeout() and move parameter checks here. In collaboration with: kib MFC after: 1 week
2011-12-03 12:30:58 +00:00
error = umtx_copyin_timeout(uap->uaddr2, &timeout);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
if (error != 0)
return (error);
ts = &timeout;
}
if a thread blocked on userland condition variable is pthread_cancel()ed, it is expected that the thread will not consume a pthread_cond_signal(), therefor, we use thr_wake() to mark a flag, the flag tells a thread calling do_cv_wait() in umtx code to not block on a condition variable. Thread library is expected that once a thread detected itself is in pthread_cond_wait, it will call the thr_wake() for itself in its SIGCANCEL handler.
2006-12-04 14:15:12 +00:00
return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val));
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
}
static int
__umtx_op_cv_signal(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_cv_signal(td, uap->obj));
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
}
static int
__umtx_op_cv_broadcast(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_cv_broadcast(td, uap->obj));
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
static int
__umtx_op_rw_rdlock(struct thread *td, struct _umtx_op_args *uap)
{
Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall.
2012-02-27 13:38:52 +00:00
struct _umtx_time timeout;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL) {
error = do_rw_rdlock(td, uap->obj, uap->val, 0);
} else {
Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall.
2012-02-27 13:38:52 +00:00
error = umtx_copyin_umtx_time(uap->uaddr2,
(size_t)uap->uaddr1, &timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
if (error != 0)
return (error);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = do_rw_rdlock(td, uap->obj, uap->val, &timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
return (error);
}
static int
__umtx_op_rw_wrlock(struct thread *td, struct _umtx_op_args *uap)
{
Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall.
2012-02-27 13:38:52 +00:00
struct _umtx_time timeout;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL) {
error = do_rw_wrlock(td, uap->obj, 0);
} else {
Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall.
2012-02-27 13:38:52 +00:00
error = umtx_copyin_umtx_time(uap->uaddr2,
(size_t)uap->uaddr1, &timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
if (error != 0)
return (error);
Merge umtxq_sleep and umtxq_nanosleep into a single function by using an abs_timeout structure which describes timeout info.
2012-03-30 05:40:26 +00:00
error = do_rw_wrlock(td, uap->obj, &timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
return (error);
}
static int
__umtx_op_rw_unlock(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_rw_unlock(td, uap->obj));
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#if defined(COMPAT_FREEBSD9) || defined(COMPAT_FREEBSD10)
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
static int
__umtx_op_sem_wait(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time(
uap->uaddr2, (size_t)uap->uaddr1, &timeout);
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
}
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
return (do_sem_wait(td, uap->obj, tm_p));
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
}
static int
__umtx_op_sem_wake(struct thread *td, struct _umtx_op_args *uap)
{
Minor (and incomplete) style cleanup. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 20:47:42 +00:00
return (do_sem_wake(td, uap->obj));
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#endif
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
static int
__umtx_op_wake2_umutex(struct thread *td, struct _umtx_op_args *uap)
{
Minor (and incomplete) style cleanup. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 20:47:42 +00:00
return (do_wake2_umutex(td, uap->obj, uap->val));
umtx operation UMTX_OP_MUTEX_WAKE has a side-effect that it accesses a mutex after a thread has unlocked it, it event writes data to the mutex memory to clear contention bit, there is a race that other threads can lock it and unlock it, then destroy it, so it should not write data to the mutex memory if there isn't any waiter. The new operation UMTX_OP_MUTEX_WAKE2 try to fix the problem. It requires thread library to clear the lock word entirely, then call the WAKE2 operation to check if there is any waiter in kernel, and try to wake up a thread, if necessary, the contention bit is set again by the operation. This also mitgates the chance that other threads find the contention bit and try to enter kernel to compete with each other to wake up sleeping thread, this is unnecessary. With this change, the mutex owner is no longer holding the mutex until it reaches a point where kernel umtx queue is locked, it releases the mutex as soon as possible. Performance is improved when the mutex is contensted heavily. On Intel i3-2310M, the runtime of a benchmark program is reduced from 26.87 seconds to 2.39 seconds, it even is better than UMTX_OP_MUTEX_WAKE which is deprecated now. http://people.freebsd.org/~davidxu/bench/mutex_perf.c
2012-04-05 02:24:08 +00:00
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
static int
__umtx_op_sem2_wait(struct thread *td, struct _umtx_op_args *uap)
{
struct _umtx_time *tm_p, timeout;
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
size_t uasize;
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
int error;
/* Allow a null timespec (wait forever). */
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
if (uap->uaddr2 == NULL) {
uasize = 0;
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
tm_p = NULL;
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
} else {
uasize = (size_t)uap->uaddr1;
error = umtx_copyin_umtx_time(uap->uaddr2, uasize, &timeout);
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
if (error != 0)
return (error);
tm_p = &timeout;
}
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
error = do_sem2_wait(td, uap->obj, tm_p);
if (error == EINTR && uap->uaddr2 != NULL &&
(timeout._flags & UMTX_ABSTIME) == 0 &&
uasize >= sizeof(struct _umtx_time) + sizeof(struct timespec)) {
error = copyout(&timeout._timeout,
(struct _umtx_time *)uap->uaddr2 + 1,
sizeof(struct timespec));
if (error == 0) {
error = EINTR;
}
}
return (error);
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
}
static int
__umtx_op_sem2_wake(struct thread *td, struct _umtx_op_args *uap)
{
Minor (and incomplete) style cleanup. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 20:47:42 +00:00
return (do_sem2_wake(td, uap->obj));
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
}
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
#define USHM_OBJ_UMTX(o) \
((struct umtx_shm_obj_list *)(&(o)->umtx_data))
#define USHMF_REG_LINKED 0x0001
#define USHMF_OBJ_LINKED 0x0002
struct umtx_shm_reg {
TAILQ_ENTRY(umtx_shm_reg) ushm_reg_link;
LIST_ENTRY(umtx_shm_reg) ushm_obj_link;
struct umtx_key ushm_key;
struct ucred *ushm_cred;
struct shmfd *ushm_obj;
u_int ushm_refcnt;
u_int ushm_flags;
};
LIST_HEAD(umtx_shm_obj_list, umtx_shm_reg);
TAILQ_HEAD(umtx_shm_reg_head, umtx_shm_reg);
static uma_zone_t umtx_shm_reg_zone;
static struct umtx_shm_reg_head umtx_shm_registry[UMTX_CHAINS];
static struct mtx umtx_shm_lock;
static struct umtx_shm_reg_head umtx_shm_reg_delfree =
TAILQ_HEAD_INITIALIZER(umtx_shm_reg_delfree);
static void umtx_shm_free_reg(struct umtx_shm_reg *reg);
static void
umtx_shm_reg_delfree_tq(void *context __unused, int pending __unused)
{
struct umtx_shm_reg_head d;
struct umtx_shm_reg *reg, *reg1;
TAILQ_INIT(&d);
mtx_lock(&umtx_shm_lock);
TAILQ_CONCAT(&d, &umtx_shm_reg_delfree, ushm_reg_link);
mtx_unlock(&umtx_shm_lock);
TAILQ_FOREACH_SAFE(reg, &d, ushm_reg_link, reg1) {
TAILQ_REMOVE(&d, reg, ushm_reg_link);
umtx_shm_free_reg(reg);
}
}
static struct task umtx_shm_reg_delfree_task =
TASK_INITIALIZER(0, umtx_shm_reg_delfree_tq, NULL);
static struct umtx_shm_reg *
umtx_shm_find_reg_locked(const struct umtx_key *key)
{
struct umtx_shm_reg *reg;
struct umtx_shm_reg_head *reg_head;
KASSERT(key->shared, ("umtx_p_find_rg: private key"));
mtx_assert(&umtx_shm_lock, MA_OWNED);
reg_head = &umtx_shm_registry[key->hash];
TAILQ_FOREACH(reg, reg_head, ushm_reg_link) {
KASSERT(reg->ushm_key.shared,
("non-shared key on reg %p %d", reg, reg->ushm_key.shared));
if (reg->ushm_key.info.shared.object ==
key->info.shared.object &&
reg->ushm_key.info.shared.offset ==
key->info.shared.offset) {
KASSERT(reg->ushm_key.type == TYPE_SHM, ("TYPE_USHM"));
KASSERT(reg->ushm_refcnt > 0,
("reg %p refcnt 0 onlist", reg));
KASSERT((reg->ushm_flags & USHMF_REG_LINKED) != 0,
("reg %p not linked", reg));
reg->ushm_refcnt++;
return (reg);
}
}
return (NULL);
}
static struct umtx_shm_reg *
umtx_shm_find_reg(const struct umtx_key *key)
{
struct umtx_shm_reg *reg;
mtx_lock(&umtx_shm_lock);
reg = umtx_shm_find_reg_locked(key);
mtx_unlock(&umtx_shm_lock);
return (reg);
}
static void
umtx_shm_free_reg(struct umtx_shm_reg *reg)
{
chgumtxcnt(reg->ushm_cred->cr_ruidinfo, -1, 0);
crfree(reg->ushm_cred);
shm_drop(reg->ushm_obj);
uma_zfree(umtx_shm_reg_zone, reg);
}
static bool
umtx_shm_unref_reg_locked(struct umtx_shm_reg *reg, bool force)
{
bool res;
mtx_assert(&umtx_shm_lock, MA_OWNED);
KASSERT(reg->ushm_refcnt > 0, ("ushm_reg %p refcnt 0", reg));
reg->ushm_refcnt--;
res = reg->ushm_refcnt == 0;
if (res || force) {
if ((reg->ushm_flags & USHMF_REG_LINKED) != 0) {
TAILQ_REMOVE(&umtx_shm_registry[reg->ushm_key.hash],
reg, ushm_reg_link);
reg->ushm_flags &= ~USHMF_REG_LINKED;
}
if ((reg->ushm_flags & USHMF_OBJ_LINKED) != 0) {
LIST_REMOVE(reg, ushm_obj_link);
reg->ushm_flags &= ~USHMF_OBJ_LINKED;
}
}
return (res);
}
static void
umtx_shm_unref_reg(struct umtx_shm_reg *reg, bool force)
{
vm_object_t object;
bool dofree;
if (force) {
object = reg->ushm_obj->shm_object;
VM_OBJECT_WLOCK(object);
object->flags |= OBJ_UMTXDEAD;
VM_OBJECT_WUNLOCK(object);
}
mtx_lock(&umtx_shm_lock);
dofree = umtx_shm_unref_reg_locked(reg, force);
mtx_unlock(&umtx_shm_lock);
if (dofree)
umtx_shm_free_reg(reg);
}
void
umtx_shm_object_init(vm_object_t object)
{
LIST_INIT(USHM_OBJ_UMTX(object));
}
void
umtx_shm_object_terminated(vm_object_t object)
{
struct umtx_shm_reg *reg, *reg1;
bool dofree;
umtx: avoid umtxshm locking on object termination if possible Sample build world result on tmpfs: kern.ipc.umtx_terminate_notempty: 0 kern.ipc.umtx_terminate_empty: 2891815 Sponsored by: The FreeBSD Foundation
2018-12-08 14:04:57 +00:00
if (LIST_EMPTY(USHM_OBJ_UMTX(object)))
return;
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
dofree = false;
mtx_lock(&umtx_shm_lock);
LIST_FOREACH_SAFE(reg, USHM_OBJ_UMTX(object), ushm_obj_link, reg1) {
if (umtx_shm_unref_reg_locked(reg, true)) {
TAILQ_INSERT_TAIL(&umtx_shm_reg_delfree, reg,
ushm_reg_link);
dofree = true;
}
}
mtx_unlock(&umtx_shm_lock);
if (dofree)
taskqueue_enqueue(taskqueue_thread, &umtx_shm_reg_delfree_task);
}
static int
umtx_shm_create_reg(struct thread *td, const struct umtx_key *key,
struct umtx_shm_reg **res)
{
struct umtx_shm_reg *reg, *reg1;
struct ucred *cred;
int error;
reg = umtx_shm_find_reg(key);
if (reg != NULL) {
*res = reg;
return (0);
}
cred = td->td_ucred;
if (!chgumtxcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_UMTXP)))
return (ENOMEM);
reg = uma_zalloc(umtx_shm_reg_zone, M_WAITOK | M_ZERO);
reg->ushm_refcnt = 1;
bcopy(key, &reg->ushm_key, sizeof(*key));
reg->ushm_obj = shm_alloc(td->td_ucred, O_RDWR);
reg->ushm_cred = crhold(cred);
error = shm_dotruncate(reg->ushm_obj, PAGE_SIZE);
if (error != 0) {
umtx_shm_free_reg(reg);
return (error);
}
mtx_lock(&umtx_shm_lock);
reg1 = umtx_shm_find_reg_locked(key);
if (reg1 != NULL) {
mtx_unlock(&umtx_shm_lock);
umtx_shm_free_reg(reg);
*res = reg1;
return (0);
}
reg->ushm_refcnt++;
TAILQ_INSERT_TAIL(&umtx_shm_registry[key->hash], reg, ushm_reg_link);
LIST_INSERT_HEAD(USHM_OBJ_UMTX(key->info.shared.object), reg,
ushm_obj_link);
reg->ushm_flags = USHMF_REG_LINKED | USHMF_OBJ_LINKED;
mtx_unlock(&umtx_shm_lock);
*res = reg;
return (0);
}
static int
umtx_shm_alive(struct thread *td, void *addr)
{
vm_map_t map;
vm_map_entry_t entry;
vm_object_t object;
vm_pindex_t pindex;
vm_prot_t prot;
int res, ret;
boolean_t wired;
map = &td->td_proc->p_vmspace->vm_map;
res = vm_map_lookup(&map, (uintptr_t)addr, VM_PROT_READ, &entry,
&object, &pindex, &prot, &wired);
if (res != KERN_SUCCESS)
return (EFAULT);
if (object == NULL)
ret = EINVAL;
else
ret = (object->flags & OBJ_UMTXDEAD) != 0 ? ENOTTY : 0;
vm_map_lookup_done(map, entry);
return (ret);
}
static void
umtx_shm_init(void)
{
int i;
umtx_shm_reg_zone = uma_zcreate("umtx_shm", sizeof(struct umtx_shm_reg),
NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
mtx_init(&umtx_shm_lock, "umtxshm", NULL, MTX_DEF);
for (i = 0; i < nitems(umtx_shm_registry); i++)
TAILQ_INIT(&umtx_shm_registry[i]);
}
static int
umtx_shm(struct thread *td, void *addr, u_int flags)
{
struct umtx_key key;
struct umtx_shm_reg *reg;
struct file *fp;
int error, fd;
if (__bitcount(flags & (UMTX_SHM_CREAT | UMTX_SHM_LOOKUP |
UMTX_SHM_DESTROY| UMTX_SHM_ALIVE)) != 1)
return (EINVAL);
if ((flags & UMTX_SHM_ALIVE) != 0)
return (umtx_shm_alive(td, addr));
error = umtx_key_get(addr, TYPE_SHM, PROCESS_SHARE, &key);
if (error != 0)
return (error);
KASSERT(key.shared == 1, ("non-shared key"));
if ((flags & UMTX_SHM_CREAT) != 0) {
error = umtx_shm_create_reg(td, &key, &reg);
} else {
reg = umtx_shm_find_reg(&key);
if (reg == NULL)
error = ESRCH;
}
umtx_key_release(&key);
if (error != 0)
return (error);
KASSERT(reg != NULL, ("no reg"));
if ((flags & UMTX_SHM_DESTROY) != 0) {
umtx_shm_unref_reg(reg, true);
} else {
#if 0
#ifdef MAC
error = mac_posixshm_check_open(td->td_ucred,
reg->ushm_obj, FFLAGS(O_RDWR));
if (error == 0)
#endif
error = shm_access(reg->ushm_obj, td->td_ucred,
FFLAGS(O_RDWR));
if (error == 0)
#endif
error = falloc_caps(td, &fp, &fd, O_CLOEXEC, NULL);
if (error == 0) {
shm_hold(reg->ushm_obj);
finit(fp, FFLAGS(O_RDWR), DTYPE_SHM, reg->ushm_obj,
&shm_ops);
td->td_retval[0] = fd;
fdrop(fp, td);
}
}
umtx_shm_unref_reg(reg, false);
return (error);
}
static int
__umtx_op_shm(struct thread *td, struct _umtx_op_args *uap)
{
return (umtx_shm(td, uap->uaddr1, uap->val));
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
static int
umtx_robust_lists(struct thread *td, struct umtx_robust_lists_params *rbp)
{
td->td_rb_list = rbp->robust_list_offset;
td->td_rbp_list = rbp->robust_priv_list_offset;
td->td_rb_inact = rbp->robust_inact_offset;
return (0);
}
static int
__umtx_op_robust_lists(struct thread *td, struct _umtx_op_args *uap)
{
struct umtx_robust_lists_params rb;
int error;
if (uap->val > sizeof(rb))
return (EINVAL);
bzero(&rb, sizeof(rb));
error = copyin(uap->uaddr1, &rb, uap->val);
if (error != 0)
return (error);
return (umtx_robust_lists(td, &rb));
}
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
typedef int (*_umtx_op_func)(struct thread *td, struct _umtx_op_args *uap);
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
static const _umtx_op_func op_table[] = {
[UMTX_OP_RESERVED0] = __umtx_op_unimpl,
[UMTX_OP_RESERVED1] = __umtx_op_unimpl,
[UMTX_OP_WAIT] = __umtx_op_wait,
[UMTX_OP_WAKE] = __umtx_op_wake,
[UMTX_OP_MUTEX_TRYLOCK] = __umtx_op_trylock_umutex,
[UMTX_OP_MUTEX_LOCK] = __umtx_op_lock_umutex,
[UMTX_OP_MUTEX_UNLOCK] = __umtx_op_unlock_umutex,
[UMTX_OP_SET_CEILING] = __umtx_op_set_ceiling,
[UMTX_OP_CV_WAIT] = __umtx_op_cv_wait,
[UMTX_OP_CV_SIGNAL] = __umtx_op_cv_signal,
[UMTX_OP_CV_BROADCAST] = __umtx_op_cv_broadcast,
[UMTX_OP_WAIT_UINT] = __umtx_op_wait_uint,
[UMTX_OP_RW_RDLOCK] = __umtx_op_rw_rdlock,
[UMTX_OP_RW_WRLOCK] = __umtx_op_rw_wrlock,
[UMTX_OP_RW_UNLOCK] = __umtx_op_rw_unlock,
[UMTX_OP_WAIT_UINT_PRIVATE] = __umtx_op_wait_uint_private,
[UMTX_OP_WAKE_PRIVATE] = __umtx_op_wake_private,
[UMTX_OP_MUTEX_WAIT] = __umtx_op_wait_umutex,
[UMTX_OP_MUTEX_WAKE] = __umtx_op_wake_umutex,
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#if defined(COMPAT_FREEBSD9) || defined(COMPAT_FREEBSD10)
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
[UMTX_OP_SEM_WAIT] = __umtx_op_sem_wait,
[UMTX_OP_SEM_WAKE] = __umtx_op_sem_wake,
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#else
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
[UMTX_OP_SEM_WAIT] = __umtx_op_unimpl,
[UMTX_OP_SEM_WAKE] = __umtx_op_unimpl,
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#endif
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
[UMTX_OP_NWAKE_PRIVATE] = __umtx_op_nwake_private,
[UMTX_OP_MUTEX_WAKE2] = __umtx_op_wake2_umutex,
[UMTX_OP_SEM2_WAIT] = __umtx_op_sem2_wait,
[UMTX_OP_SEM2_WAKE] = __umtx_op_sem2_wake,
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
[UMTX_OP_SHM] = __umtx_op_shm,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
[UMTX_OP_ROBUST_LISTS] = __umtx_op_robust_lists,
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
};
int
In order to maximize the re-usability of kernel code in user space this patch modifies makesyscalls.sh to prefix all of the non-compatibility calls (e.g. not linux_, freebsd32_) with sys_ and updates the kernel entry points and all places in the code that use them. It also fixes an additional name space collision between the kernel function psignal and the libc function of the same name by renaming the kernel psignal kern_psignal(). By introducing this change now we will ease future MFCs that change syscalls. Reviewed by: rwatson Approved by: re (bz)
2011-09-16 13:58:51 +00:00
sys__umtx_op(struct thread *td, struct _umtx_op_args *uap)
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
{
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
if ((unsigned)uap->op < nitems(op_table))
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
return (*op_table[uap->op])(td, uap);
return (EINVAL);
}
Provide groundwork for 32-bit binary compatibility on non-x86 platforms, for upcoming 64-bit PowerPC and MIPS support. This renames the COMPAT_IA32 option to COMPAT_FREEBSD32, removes some IA32-specific code from MI parts of the kernel and enhances the freebsd32 compatibility code to support big-endian platforms. Reviewed by: kib, jhb
2010-03-11 14:49:06 +00:00
#ifdef COMPAT_FREEBSD32
Implement 32bit umtx_lock and umtx_unlock system calls, these two system calls are not used by libthr in RELENG_6 and HEAD, it is only used by the libthr in RELENG-5, the _umtx_op system call can do more incremental dirty works than these two system calls without having to introduce new system calls or throw away old system calls when things are going on.
2006-10-06 08:22:08 +00:00
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
struct timespec32 {
The fields of struct timespec32 should be int32_t and not uint32_t. Make this change. Reviewed by: bde, davidxu Tested by: pho MFC after: 1 week
2012-10-27 23:42:41 +00:00
int32_t tv_sec;
int32_t tv_nsec;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
};
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct umtx_time32 {
struct timespec32 timeout;
uint32_t flags;
uint32_t clockid;
};
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
static inline int
Rename copyin_timeout32 to umtx_copyin_timeout32 and move parameter check here. Include check for negative seconds value. In collaboration with: kib MFC after: 1 week
2011-12-03 12:28:33 +00:00
umtx_copyin_timeout32(void *addr, struct timespec *tsp)
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
{
struct timespec32 ts32;
int error;
error = copyin(addr, &ts32, sizeof(struct timespec32));
if (error == 0) {
Rename copyin_timeout32 to umtx_copyin_timeout32 and move parameter check here. Include check for negative seconds value. In collaboration with: kib MFC after: 1 week
2011-12-03 12:28:33 +00:00
if (ts32.tv_sec < 0 ||
ts32.tv_nsec >= 1000000000 ||
ts32.tv_nsec < 0)
error = EINVAL;
else {
tsp->tv_sec = ts32.tv_sec;
tsp->tv_nsec = ts32.tv_nsec;
}
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
Revert my previous errno hack, that is certainly an issue, and always has been, but the system call itself returns errno in a register so the problem is really a function of libc, not the system call. Discussed with : Matthew Dillion <dillon@apollo.backplane.com>
2005-01-18 13:53:10 +00:00
return (error);
1. make umtx sharable between processes, the way is two or more processes call mmap() to create a shared space, and then initialize umtx on it, after that, each thread in different processes can use the umtx same as threads in same process. 2. introduce a new syscall _umtx_op to support timed lock and condition variable semantics. also, orignal umtx_lock and umtx_unlock inline functions now are reimplemented by using _umtx_op, the _umtx_op can use arbitrary id not just a thread id.
2004-12-18 12:52:44 +00:00
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
static inline int
umtx_copyin_umtx_time32(const void *addr, size_t size, struct _umtx_time *tp)
{
struct umtx_time32 t32;
int error;
t32.clockid = CLOCK_REALTIME;
t32.flags = 0;
if (size <= sizeof(struct timespec32))
error = copyin(addr, &t32.timeout, sizeof(struct timespec32));
else
error = copyin(addr, &t32, sizeof(struct umtx_time32));
if (error != 0)
return (error);
if (t32.timeout.tv_sec < 0 ||
t32.timeout.tv_nsec >= 1000000000 || t32.timeout.tv_nsec < 0)
return (EINVAL);
tp->_timeout.tv_sec = t32.timeout.tv_sec;
tp->_timeout.tv_nsec = t32.timeout.tv_nsec;
tp->_flags = t32.flags;
tp->_clockid = t32.clockid;
return (0);
}
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
static int
__umtx_op_wait_compat32(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
int error;
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time32(uap->uaddr2,
(size_t)uap->uaddr1, &timeout);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_wait(td, uap->obj, uap->val, tm_p, 1, 0));
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
static int
__umtx_op_lock_umutex_compat32(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
else {
Use umtx_copyin_umtx_time32() in __umtx_op_lock_umutex_compat32(). Non-NULL timeouts where copied in improperly and could produce failures due to incompatible data structures. Reviewed by: kib MFC after: 3 days Sponsored by: DARPA, AFRL Differential Revision: https://reviews.freebsd.org/D14587
2018-03-06 01:52:04 +00:00
error = umtx_copyin_umtx_time32(uap->uaddr2,
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
(size_t)uap->uaddr1, &timeout);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_lock_umutex(td, uap->obj, tm_p, 0));
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
static int
__umtx_op_wait_umutex_compat32(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time32(uap->uaddr2,
(size_t)uap->uaddr1, &timeout);
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_lock_umutex(td, uap->obj, tm_p, _UMUTEX_WAIT));
Add two commands to _umtx_op system call to allow a simple mutex to be locked and unlocked completely in userland. by locking and unlocking mutex in userland, it reduces the total time a mutex is locked by a thread, in some application code, a mutex only protects a small piece of code, the code's execution time is less than a simple system call, if a lock contention happens, however in current implemenation, the lock holder has to extend its locking time and enter kernel to unlock it, the change avoids this disadvantage, it first sets mutex to free state and then enters kernel and wake one waiter up. This improves performance dramatically in some sysbench mutex tests. Tested by: kris Sounds great: jeff
2008-06-24 07:32:12 +00:00
}
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
static int
__umtx_op_cv_wait_compat32(struct thread *td, struct _umtx_op_args *uap)
{
struct timespec *ts, timeout;
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL)
ts = NULL;
else {
Rename copyin_timeout32 to umtx_copyin_timeout32 and move parameter check here. Include check for negative seconds value. In collaboration with: kib MFC after: 1 week
2011-12-03 12:28:33 +00:00
error = umtx_copyin_timeout32(uap->uaddr2, &timeout);
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
if (error != 0)
return (error);
ts = &timeout;
}
if a thread blocked on userland condition variable is pthread_cancel()ed, it is expected that the thread will not consume a pthread_cond_signal(), therefor, we use thr_wake() to mark a flag, the flag tells a thread calling do_cv_wait() in umtx code to not block on a condition variable. Thread library is expected that once a thread detected itself is in pthread_cond_wait, it will call the thr_wake() for itself in its SIGCANCEL handler.
2006-12-04 14:15:12 +00:00
return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val));
Introduce userspace condition variable, since we have already POSIX priority mutex implemented, it is the time to introduce this stuff, now we can use umutex and ucond together to implement pthread's condition wait/signal.
2006-12-03 01:49:22 +00:00
}
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
static int
__umtx_op_rw_rdlock_compat32(struct thread *td, struct _umtx_op_args *uap)
{
Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall.
2012-02-27 13:38:52 +00:00
struct _umtx_time timeout;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL) {
error = do_rw_rdlock(td, uap->obj, uap->val, 0);
} else {
Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall.
2012-02-27 13:38:52 +00:00
error = umtx_copyin_umtx_time32(uap->uaddr2,
(size_t)uap->uaddr1, &timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
if (error != 0)
return (error);
Fix COMPAT_FREEBSD32 build. Submitted by: Andreas Tobler < andreast at fgznet dot ch >
2012-03-30 09:03:53 +00:00
error = do_rw_rdlock(td, uap->obj, uap->val, &timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
return (error);
}
static int
__umtx_op_rw_wrlock_compat32(struct thread *td, struct _umtx_op_args *uap)
{
Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall.
2012-02-27 13:38:52 +00:00
struct _umtx_time timeout;
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL) {
Fix compiling problem for amd64.
2008-04-02 05:54:41 +00:00
error = do_rw_wrlock(td, uap->obj, 0);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
} else {
Follow changes made in revision 232144, pass absolute timeout to kernel, this eliminates a clock_gettime() syscall.
2012-02-27 13:38:52 +00:00
error = umtx_copyin_umtx_time32(uap->uaddr2,
(size_t)uap->uaddr1, &timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
if (error != 0)
return (error);
Fix COMPAT_FREEBSD32 build. Submitted by: Andreas Tobler < andreast at fgznet dot ch >
2012-03-30 09:03:53 +00:00
error = do_rw_wrlock(td, uap->obj, &timeout);
Introduce kernel based userland rwlock. Each umtx chain now has two lists, one for readers and one for writers, other types of synchronization object just use first list. Asked by: jeff
2008-04-02 04:08:37 +00:00
}
return (error);
}
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
static int
__umtx_op_wait_uint_private_compat32(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
int error;
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time32(
uap->uaddr2, (size_t)uap->uaddr1,&timeout);
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
return (do_wait(td, uap->obj, uap->val, tm_p, 1, 1));
Introduce command UMTX_OP_WAIT_UINT_PRIVATE and UMTX_OP_WAKE_PRIVATE to allow userland to specify that an address is not shared by multiple processes.
2008-04-29 03:48:48 +00:00
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#if defined(COMPAT_FREEBSD9) || defined(COMPAT_FREEBSD10)
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
static int
__umtx_op_sem_wait_compat32(struct thread *td, struct _umtx_op_args *uap)
{
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
struct _umtx_time *tm_p, timeout;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
int error;
/* Allow a null timespec (wait forever). */
if (uap->uaddr2 == NULL)
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = NULL;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
else {
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
error = umtx_copyin_umtx_time32(uap->uaddr2,
(size_t)uap->uaddr1, &timeout);
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
if (error != 0)
return (error);
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
tm_p = &timeout;
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
}
In revision 231989, we pass a 16-bit clock ID into kernel, however according to POSIX document, the clock ID may be dynamically allocated, it unlikely will be in 64K forever. To make it future compatible, we pack all timeout information into a new structure called _umtx_time, and use fourth argument as a size indication, a zero means it is old code using timespec as timeout value, but the new structure also includes flags and a clock ID, so the size argument is different than before, and it is non-zero. With this change, it is possible that a thread can sleep on any supported clock, though current kernel code does not have such a POSIX clock driver system.
2012-02-25 02:12:17 +00:00
return (do_sem_wait(td, uap->obj, tm_p));
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
}
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#endif
static int
__umtx_op_sem2_wait_compat32(struct thread *td, struct _umtx_op_args *uap)
{
struct _umtx_time *tm_p, timeout;
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
size_t uasize;
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
int error;
/* Allow a null timespec (wait forever). */
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
if (uap->uaddr2 == NULL) {
uasize = 0;
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
tm_p = NULL;
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
} else {
uasize = (size_t)uap->uaddr1;
error = umtx_copyin_umtx_time32(uap->uaddr2, uasize, &timeout);
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
if (error != 0)
return (error);
tm_p = &timeout;
}
Add sem_clockwait_np() This function allows the caller to specify the reference clock and choose between absolute and relative mode. In relative mode, the remaining time can be returned. The API is similar to clock_nanosleep(3). Thanks to Ed Schouten for that suggestion. While I'm here, reduce the sleep time in the semaphore "child" test to greatly reduce its runtime. Also add a reasonable timeout. Reviewed by: ed (userland) MFC after: 2 weeks Relnotes: yes Sponsored by: Dell EMC Differential Revision: https://reviews.freebsd.org/D9656
2017-02-23 19:36:38 +00:00
error = do_sem2_wait(td, uap->obj, tm_p);
if (error == EINTR && uap->uaddr2 != NULL &&
(timeout._flags & UMTX_ABSTIME) == 0 &&
uasize >= sizeof(struct umtx_time32) + sizeof(struct timespec32)) {
struct timespec32 remain32 = {
.tv_sec = timeout._timeout.tv_sec,
.tv_nsec = timeout._timeout.tv_nsec
};
error = copyout(&remain32,
(struct umtx_time32 *)uap->uaddr2 + 1,
sizeof(struct timespec32));
if (error == 0) {
error = EINTR;
}
}
return (error);
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
}
Add user-level semaphore synchronous type, this change allows multiple processes to share semaphore by using shared memory area, in simplest case, only one atomic operation is needed in userland, waiter flag is maintained by kernel and userland only checks the flag, if the flag is set, user code enters kernel and does a wakeup() call. Move type definitions into file _umtx.h to minimize compiling time. Also type names need to be prefixed with underline character, this would reduce name conflict (still in progress).
2010-01-04 05:27:49 +00:00
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
static int
__umtx_op_nwake_private32(struct thread *td, struct _umtx_op_args *uap)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uint32_t uaddrs[BATCH_SIZE], **upp;
int count, error, i, pos, tocopy;
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
upp = (uint32_t **)uap->obj;
error = 0;
for (count = uap->val, pos = 0; count > 0; count -= tocopy,
pos += tocopy) {
tocopy = MIN(count, BATCH_SIZE);
error = copyin(upp + pos, uaddrs, tocopy * sizeof(uint32_t));
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
if (error != 0)
break;
for (i = 0; i < tocopy; ++i)
kern_umtx_wake(td, (void *)(intptr_t)uaddrs[i],
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
INT_MAX, 1);
maybe_yield();
MFp4: - Add flags CVWAIT_ABSTIME and CVWAIT_CLOCKID for umtx kernel based condition variable, this should eliminate an extra system call to get current time. - Add sub-function UMTX_OP_NWAKE_PRIVATE to wake up N channels in single system call. Create userland sleep queue for condition variable, in most cases, thread will wait in the queue, the pthread_cond_signal will defer thread wakeup until the mutex is unlocked, it tries to avoid an extra system call and a extra context switch in time window of pthread_cond_signal and pthread_mutex_unlock. The changes are part of process-shared mutex project.
2010-12-22 05:01:52 +00:00
}
return (error);
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
struct umtx_robust_lists_params_compat32 {
uint32_t robust_list_offset;
uint32_t robust_priv_list_offset;
uint32_t robust_inact_offset;
};
static int
__umtx_op_robust_lists_compat32(struct thread *td, struct _umtx_op_args *uap)
{
struct umtx_robust_lists_params rb;
struct umtx_robust_lists_params_compat32 rb32;
int error;
if (uap->val > sizeof(rb32))
return (EINVAL);
bzero(&rb, sizeof(rb));
bzero(&rb32, sizeof(rb32));
error = copyin(uap->uaddr1, &rb32, uap->val);
if (error != 0)
return (error);
rb.robust_list_offset = rb32.robust_list_offset;
rb.robust_priv_list_offset = rb32.robust_priv_list_offset;
rb.robust_inact_offset = rb32.robust_inact_offset;
return (umtx_robust_lists(td, &rb));
}
Also mark compat32 umtx op table as constant. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:32:30 +00:00
static const _umtx_op_func op_table_compat32[] = {
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
[UMTX_OP_RESERVED0] = __umtx_op_unimpl,
[UMTX_OP_RESERVED1] = __umtx_op_unimpl,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
[UMTX_OP_WAIT] = __umtx_op_wait_compat32,
[UMTX_OP_WAKE] = __umtx_op_wake,
Fix umtx lock/trylock for compat32. Sponsored by: The FreeBSD Foundation MFC after: 3 days
2016-04-19 11:37:43 +00:00
[UMTX_OP_MUTEX_TRYLOCK] = __umtx_op_trylock_umutex,
[UMTX_OP_MUTEX_LOCK] = __umtx_op_lock_umutex_compat32,
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
[UMTX_OP_MUTEX_UNLOCK] = __umtx_op_unlock_umutex,
[UMTX_OP_SET_CEILING] = __umtx_op_set_ceiling,
[UMTX_OP_CV_WAIT] = __umtx_op_cv_wait_compat32,
[UMTX_OP_CV_SIGNAL] = __umtx_op_cv_signal,
[UMTX_OP_CV_BROADCAST] = __umtx_op_cv_broadcast,
[UMTX_OP_WAIT_UINT] = __umtx_op_wait_compat32,
[UMTX_OP_RW_RDLOCK] = __umtx_op_rw_rdlock_compat32,
[UMTX_OP_RW_WRLOCK] = __umtx_op_rw_wrlock_compat32,
[UMTX_OP_RW_UNLOCK] = __umtx_op_rw_unlock,
[UMTX_OP_WAIT_UINT_PRIVATE] = __umtx_op_wait_uint_private_compat32,
[UMTX_OP_WAKE_PRIVATE] = __umtx_op_wake_private,
[UMTX_OP_MUTEX_WAIT] = __umtx_op_wait_umutex_compat32,
[UMTX_OP_MUTEX_WAKE] = __umtx_op_wake_umutex,
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#if defined(COMPAT_FREEBSD9) || defined(COMPAT_FREEBSD10)
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
[UMTX_OP_SEM_WAIT] = __umtx_op_sem_wait_compat32,
[UMTX_OP_SEM_WAKE] = __umtx_op_sem_wake,
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#else
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
[UMTX_OP_SEM_WAIT] = __umtx_op_unimpl,
[UMTX_OP_SEM_WAKE] = __umtx_op_unimpl,
The current POSIX semaphore implementation stores the _has_waiters flag in a separate word from the _count. This does not permit both items to be updated atomically in a portable manner. As a result, sem_post() must always perform a system call to safely clear _has_waiters. This change removes the _has_waiters field and instead uses the high bit of _count as the _has_waiters flag. A new umtx object type (_usem2) and two new umtx operations are added (SEM_WAIT2 and SEM_WAKE2) to implement these semantics. The older operations are still supported under the COMPAT_FREEBSD9/10 options. The POSIX semaphore API in libc has been updated to use the new implementation. Note that the new implementation is not compatible with the previous implementation. However, this only affects static binaries (which cannot be helped by symbol versioning). Binaries using a dynamic libc will continue to work fine. SEM_MAGIC has been bumped so that mismatched binaries will error rather than corrupting a shared semaphore. In addition, a padding field has been added to sem_t so that it remains the same size. Differential Revision: https://reviews.freebsd.org/D961 Reported by: adrian Reviewed by: kib, jilles (earlier version) Sponsored by: Norse
2014-10-24 20:02:44 +00:00
#endif
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
[UMTX_OP_NWAKE_PRIVATE] = __umtx_op_nwake_private32,
[UMTX_OP_MUTEX_WAKE2] = __umtx_op_wake2_umutex,
[UMTX_OP_SEM2_WAIT] = __umtx_op_sem2_wait_compat32,
[UMTX_OP_SEM2_WAKE] = __umtx_op_sem2_wake,
Implement process-shared locks support for libthr.so.3, without breaking the ABI. Special value is stored in the lock pointer to indicate shared lock, and offline page in the shared memory is allocated to store the actual lock. Reviewed by: vangyzen (previous version) Discussed with: deischen, emaste, jhb, rwatson, Martin Simmons <martin@lispworks.com> Tested by: pho Sponsored by: The FreeBSD Foundation
2016-02-28 17:52:33 +00:00
[UMTX_OP_SHM] = __umtx_op_shm,
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
[UMTX_OP_ROBUST_LISTS] = __umtx_op_robust_lists_compat32,
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
};
int
Make freebsd32_umtx_op follow the freebsd32_foo convention. Sponsored by: DARPA, AFRL
2018-11-09 00:46:10 +00:00
freebsd32__umtx_op(struct thread *td, struct freebsd32__umtx_op_args *uap)
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
{
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
if ((unsigned)uap->op < nitems(op_table_compat32)) {
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
return (*op_table_compat32[uap->op])(td,
Use C99 array initialization, which also makes the code self-documented, and eases addition of new ops. For the similar reasons, eliminate UMTX_OP_MAX. nitems() handles the only use of the symbol. Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-10-30 19:20:40 +00:00
(struct _umtx_op_args *)uap);
}
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
return (EINVAL);
}
#endif
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
void
umtx_thread_init(struct thread *td)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
td->td_umtxq = umtxq_alloc();
td->td_umtxq->uq_thread = td;
}
void
umtx_thread_fini(struct thread *td)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtxq_free(td->td_umtxq);
}
/*
* It will be called when new thread is created, e.g fork().
*/
void
umtx_thread_alloc(struct thread *td)
{
struct umtx_q *uq;
uq = td->td_umtxq;
uq->uq_inherited_pri = PRI_MAX;
KASSERT(uq->uq_flags == 0, ("uq_flags != 0"));
KASSERT(uq->uq_thread == td, ("uq_thread != td"));
KASSERT(uq->uq_pi_blocked == NULL, ("uq_pi_blocked != NULL"));
KASSERT(TAILQ_EMPTY(&uq->uq_pi_contested), ("uq_pi_contested is not empty"));
}
/*
* exec() hook.
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
*
* Clear robust lists for all process' threads, not delaying the
* cleanup to thread_exit hook, since the relevant address space is
* destroyed right now.
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
*/
static void
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
umtx_exec_hook(void *arg __unused, struct proc *p,
struct image_params *imgp __unused)
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
struct thread *td;
KASSERT(p == curproc, ("need curproc"));
KASSERT((p->p_flag & P_HADTHREADS) == 0 ||
(p->p_flag & P_STOPPED_SINGLE) != 0,
("curproc must be single-threaded"));
Reduce umtx-related work on exec and exit - there is no need to take the process lock to iterate the thread list after single-threading is enforced - typically there are no mutexes to clean up (testable without taking the global umtx lock) - typically there is no need to adjust the priority (testable without taking thread lock) Reviewed by: kib Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20160
2019-05-08 16:30:38 +00:00
/*
* There is no need to lock the list as only this thread can be
* running.
*/
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
FOREACH_THREAD_IN_PROC(p, td) {
KASSERT(td == curthread ||
((td->td_flags & TDF_BOUNDARY) != 0 && TD_IS_SUSPENDED(td)),
("running thread %p %p", p, td));
umtx_thread_cleanup(td);
td->td_rb_list = td->td_rbp_list = td->td_rb_inact = 0;
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
/*
* thread_exit() hook.
*/
void
umtx_thread_exit(struct thread *td)
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
umtx_thread_cleanup(td);
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
static int
umtx_read_uptr(struct thread *td, uintptr_t ptr, uintptr_t *res)
{
u_long res1;
#ifdef COMPAT_FREEBSD32
uint32_t res32;
#endif
int error;
#ifdef COMPAT_FREEBSD32
if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
error = fueword32((void *)ptr, &res32);
if (error == 0)
res1 = res32;
} else
#endif
{
error = fueword((void *)ptr, &res1);
}
if (error == 0)
*res = res1;
else
error = EFAULT;
return (error);
}
static void
umtx_read_rb_list(struct thread *td, struct umutex *m, uintptr_t *rb_list)
{
#ifdef COMPAT_FREEBSD32
struct umutex32 m32;
if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
memcpy(&m32, m, sizeof(m32));
*rb_list = m32.m_rb_lnk;
} else
#endif
*rb_list = m->m_rb_lnk;
}
static int
umtx_handle_rb(struct thread *td, uintptr_t rbp, uintptr_t *rb_list, bool inact)
{
struct umutex m;
int error;
KASSERT(td->td_proc == curproc, ("need current vmspace"));
error = copyin((void *)rbp, &m, sizeof(m));
if (error != 0)
return (error);
if (rb_list != NULL)
umtx_read_rb_list(td, &m, rb_list);
if ((m.m_flags & UMUTEX_ROBUST) == 0)
return (EINVAL);
if ((m.m_owner & ~UMUTEX_CONTESTED) != td->td_tid)
/* inact is cleared after unlock, allow the inconsistency */
return (inact ? 0 : EINVAL);
return (do_unlock_umutex(td, (struct umutex *)rbp, true));
}
static void
umtx_cleanup_rb_list(struct thread *td, uintptr_t rb_list, uintptr_t *rb_inact,
const char *name)
{
int error, i;
uintptr_t rbp;
bool inact;
if (rb_list == 0)
return;
error = umtx_read_uptr(td, rb_list, &rbp);
for (i = 0; error == 0 && rbp != 0 && i < umtx_max_rb; i++) {
if (rbp == *rb_inact) {
inact = true;
*rb_inact = 0;
} else
inact = false;
error = umtx_handle_rb(td, rbp, &rbp, inact);
}
if (i == umtx_max_rb && umtx_verbose_rb) {
uprintf("comm %s pid %d: reached umtx %smax rb %d\n",
td->td_proc->p_comm, td->td_proc->p_pid, name, umtx_max_rb);
}
if (error != 0 && umtx_verbose_rb) {
uprintf("comm %s pid %d: handling %srb error %d\n",
td->td_proc->p_comm, td->td_proc->p_pid, name, error);
}
}
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
/*
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
* Clean up umtx data.
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
*/
static void
umtx_thread_cleanup(struct thread *td)
{
struct umtx_q *uq;
struct umtx_pi *pi;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
uintptr_t rb_inact;
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
/*
* Disown pi mutexes.
*/
uq = td->td_umtxq;
if (uq != NULL) {
Reduce umtx-related work on exec and exit - there is no need to take the process lock to iterate the thread list after single-threading is enforced - typically there are no mutexes to clean up (testable without taking the global umtx lock) - typically there is no need to adjust the priority (testable without taking thread lock) Reviewed by: kib Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20160
2019-05-08 16:30:38 +00:00
if (uq->uq_inherited_pri != PRI_MAX ||
!TAILQ_EMPTY(&uq->uq_pi_contested)) {
mtx_lock(&umtx_lock);
uq->uq_inherited_pri = PRI_MAX;
while ((pi = TAILQ_FIRST(&uq->uq_pi_contested)) != NULL) {
pi->pi_owner = NULL;
TAILQ_REMOVE(&uq->uq_pi_contested, pi, pi_link);
}
mtx_unlock(&umtx_lock);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
}
Reduce umtx-related work on exec and exit - there is no need to take the process lock to iterate the thread list after single-threading is enforced - typically there are no mutexes to clean up (testable without taking the global umtx lock) - typically there is no need to adjust the priority (testable without taking thread lock) Reviewed by: kib Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20160
2019-05-08 16:30:38 +00:00
sched_lend_user_prio_cond(td, PRI_MAX);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
Reduce umtx-related work on exec and exit - there is no need to take the process lock to iterate the thread list after single-threading is enforced - typically there are no mutexes to clean up (testable without taking the global umtx lock) - typically there is no need to adjust the priority (testable without taking thread lock) Reviewed by: kib Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20160
2019-05-08 16:30:38 +00:00
if (td->td_rb_inact == 0 && td->td_rb_list == 0 && td->td_rbp_list == 0)
return;
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
/*
* Handle terminated robust mutexes. Must be done after
* robust pi disown, otherwise unlock could see unowned
* entries.
*/
rb_inact = td->td_rb_inact;
if (rb_inact != 0)
(void)umtx_read_uptr(td, rb_inact, &rb_inact);
umtx_cleanup_rb_list(td, td->td_rb_list, &rb_inact, "");
umtx_cleanup_rb_list(td, td->td_rbp_list, &rb_inact, "priv ");
if (rb_inact != 0)
(void)umtx_handle_rb(td, rb_inact, NULL, true);
This is initial version of POSIX priority mutex support, a new userland mutex structure is added as following: struct umutex { __lwpid_t m_owner; uint32_t m_flags; uint32_t m_ceilings[2]; uint32_t m_spare[4]; }; The m_owner represents owner thread, it is a thread id, in non-contested case, userland can simply use atomic_cmpset_int to lock the mutex, if the mutex is contested, high order bit will be set, and userland should do locking and unlocking via kernel syscall. Flag UMUTEX_PRIO_INHERIT represents pthread's PTHREAD_PRIO_INHERIT mutex, which when contention happens, kernel should do priority propagating. Flag UMUTEX_PRIO_PROTECT indicates it is pthread's PTHREAD_PRIO_PROTECT mutex, userland should initialize m_owner to contested state UMUTEX_CONTESTED, then atomic_cmpset_int will be failure and kernel syscall should be invoked to do locking, this becauses for such a mutex, kernel should always boost the thread's priority before it can lock the mutex, m_ceilings is used by PTHREAD_PRIO_PROTECT mutex, the first element is used to boost thread's priority when it locked the mutex, second element is used when the mutex is unlocked, the PTHREAD_PRIO_PROTECT mutex's link list is kept in userland, the m_ceiling[1] is managed by thread library so kernel needn't allocate memory to keep the link list, when such a mutex is unlocked, kernel reset m_owner to UMUTEX_CONTESTED. Flag USYNC_PROCESS_SHARED indicate if the synchronization object is process shared, if the flag is not set, it saves a vm_map_lookup() call. The umtx chain is still used as a sleep queue, when a thread is blocked on PTHREAD_PRIO_INHERIT mutex, a umtx_pi is allocated to support priority propagating, it is dynamically allocated and reference count is used, it is not optimized but works well in my tests, while the umtx chain has its own locking protocol, the priority propagating protocol are all protected by sched_lock because priority propagating function is called with sched_lock held from scheduler. No visible performance degradation is found which these changes. Some parameter names in _umtx_op syscall are renamed.
2006-08-28 04:24:51 +00:00
}
Reference in New Issue Copy Permalink