d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
51c3f72bfa
freebsd-skq/sys/kern/kern_umtx.c

3919 lines
88 KiB
C
Raw Normal View History

/* -> /*- for copyright notices, minor format tweaks as necessary
2005-01-06 23:35:40 +00:00
/*-
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.
*
* 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 umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
#include "opt_compat.h"
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>
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>
- 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>
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>
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>
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>
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)
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
#define UMTXQ_BUSY_ASSERT(uc) KASSERT(&(uc)->uc_busy, ("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
/*
* 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
Enlarge hash table for new condition variable.
2010-12-23 03:12:03 +00:00
#define UMTX_CHAINS 512
#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;
struct timespec cur;
struct timespec end;
};
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 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
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);
static int do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags);
static void umtx_thread_cleanup(struct thread *td);
static void umtx_exec_hook(void *arg __unused, struct proc *p __unused,
struct image_params *imgp __unused);
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
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_init(&umtx_lock, "umtx lock", NULL, MTX_SPIN);
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);
}
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);
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 = malloc(sizeof(struct umtxq_queue), M_UMTX, M_WAITOK | M_ZERO);
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)
{
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
{
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
unsigned n = (uintptr_t)key->info.both.a + key->info.both.b;
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
{
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
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)
{
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(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
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)
{
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;
- 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;
uc = umtxq_getchain(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
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
}
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
umtxq_check_susp(struct thread *td)
{
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);
if (P_SHOULDSTOP(p) ||
((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND))) {
if (p->p_flag & P_SINGLE_EXIT)
error = EINTR;
else
error = ERESTART;
}
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
{
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;
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;
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(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
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)
{
struct umtxq_chain *uc;
uc = umtxq_getchain(&uq->uq_key);
UMTXQ_LOCKED_ASSERT(uc);
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) {
kern_clock_gettime(curthread, clockid, &timo->end);
timo->cur = timo->end;
timespecadd(&timo->end, timeout);
} else {
timo->end = *timeout;
kern_clock_gettime(curthread, clockid, &timo->cur);
}
}
static void
abs_timeout_init2(struct abs_timeout *timo, const struct _umtx_time *umtxtime)
{
abs_timeout_init(timo, umtxtime->_clockid,
(umtxtime->_flags & UMTX_ABSTIME) != 0,
&umtxtime->_timeout);
}
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)
{
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);
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
tts = timo->end;
timespecsub(&tts, &timo->cur);
return (tstohz(&tts));
}
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
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 (;;) {
if (!(uq->uq_flags & UQF_UMTXQ))
return (0);
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
if (abstime != NULL) {
timo = abs_timeout_gethz(abstime);
if (timo < 0)
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
return (ETIMEDOUT);
} 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);
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
if (error != EWOULDBLOCK) {
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;
}
Some style fixes inspired by @bde.
2012-08-11 23:48:39 +00:00
if (abstime != NULL)
abs_timeout_update(abstime);
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);
}
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
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_get(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) {
return EFAULT;
}
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;
key->info.shared.offset = entry->offset + entry->start -
(vm_offset_t)addr;
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
/*
* Lock a umtx object.
*/
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
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_umtx(struct thread *td, struct umtx *umtx, u_long id,
const struct timespec *timeout)
- 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
{
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 owner;
u_long old;
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
int error = 0;
- 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
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;
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_init(&timo, CLOCK_REALTIME, 0, 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
- 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
/*
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
- 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
* can fault on any access.
*/
for (;;) {
/*
* Try the uncontested case. This should be done in userland.
*/
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
owner = casuword(&umtx->u_owner, UMTX_UNOWNED, id);
- 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
/* The acquire succeeded. */
- 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
if (owner == UMTX_UNOWNED)
return (0);
- 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
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
/* The address was invalid. */
if (owner == -1)
return (EFAULT);
- 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
/* If no one owns it but it is contested try to acquire it. */
if (owner == UMTX_CONTESTED) {
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
owner = casuword(&umtx->u_owner,
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_CONTESTED, id | UMTX_CONTESTED);
- 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. 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
if (owner == UMTX_CONTESTED)
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
/* The address was invalid. */
if (owner == -1)
return (EFAULT);
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 = umtxq_check_susp(td);
if (error != 0)
break;
- 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
/* If this failed the lock has changed, restart. */
continue;
- 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
}
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
/*
* If we caught a signal, we have retried and now
* exit immediately.
*/
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 (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
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 ((error = umtx_key_get(umtx, TYPE_SIMPLE_LOCK,
AUTO_SHARE, &uq->uq_key)) != 0)
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
return (error);
- 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
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_insert(uq);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
- 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
/*
* 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.
*/
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
old = casuword(&umtx->u_owner, owner, owner | UMTX_CONTESTED);
- 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
/* The address was invalid. */
- Make casuptr return the old value of the location we're trying to update, and change the umtx code to expect this. Reviewed by: jeff
2003-04-02 08:02:27 +00:00
if (old == -1) {
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
umtxq_lock(&uq->uq_key);
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
- 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
return (EFAULT);
- 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
}
/*
- 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
* We set the contested bit, sleep. Otherwise the lock changed
Fix umtx locking, for libthr, in the kernel. 1. There was a race condition between a thread unlocking a umtx and the thread contesting it. If the unlocking thread won the race it may try to wakeup a thread that was not yet in msleep(). The contesting thread would then go to sleep to await a wakeup that would never come. It's not possible to close the race by using a lock because calls to casuptr() may have to fault a page in from swap. Instead, the race was closed by introducing a flag that the unlocking thread will set when waking up a thread. The contesting thread will check for this flag before going to sleep. For now the flag is kept in td_flags, but it may be better to use some other member or create a new one because of the possible performance/contention issues of having to own sched_lock. Thanks to jhb for pointing me in the right direction on this one. 2. Once a umtx was contested all future locks and unlocks were happening in the kernel, regardless of whether it was contested or not. To prevent this from happening, when a thread locks a umtx it checks the queue for that umtx and unsets the contested bit if there are no other threads waiting on it. Again, this is slightly more complicated than it needs to be because we can't hold a lock across casuptr(). So, the thread has to check the queue again after unseting the bit, and reset the contested bit if it finds that another thread has put itself on the queue in the mean time. 3. Remove the if... block for unlocking an uncontested umtx, and replace it with a KASSERT. The _only_ time a thread should be unlocking a umtx in the kernel is if it is contested.
2003-07-17 11:06:40 +00:00
* and we need to retry or we lost a race to the thread
* unlocking the umtx.
- 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
*/
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
umtxq_lock(&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
if (old == owner)
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, "umtx", timeout == NULL ? NULL :
&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
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)
error = umtxq_check_susp(td);
- 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
}
Move the decision on whether to unset the contested bit or not from lock to unlock time. Suggested by: jhb
2003-07-18 17:58:37 +00:00
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
if (timeout == NULL) {
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
/* Mutex locking is restarted if it is interrupted. */
if (error == EINTR)
error = ERESTART;
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
} else {
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
/* Timed-locking is not restarted. */
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
/*
* Unlock a umtx object.
*/
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_unlock_umtx(struct thread *td, struct umtx *umtx, u_long id)
- 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
{
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;
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 owner;
u_long old;
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 error;
int count;
- 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
/*
* Make sure we own this mtx.
*/
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
owner = fuword(__DEVOLATILE(u_long *, &umtx->u_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 (owner == -1)
- 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
return (EFAULT);
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
if ((owner & ~UMTX_CONTESTED) != id)
- 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
return (EPERM);
- 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
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 & UMTX_CONTESTED) == 0) {
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
old = casuword(&umtx->u_owner, owner, UMTX_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
if (old == -1)
return (EFAULT);
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
}
- 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
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(umtx, TYPE_SIMPLE_LOCK, 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 (error);
umtxq_lock(&key);
umtxq_busy(&key);
count = umtxq_count(&key);
umtxq_unlock(&key);
Move the decision on whether to unset the contested bit or not from lock to unlock time. Suggested by: jhb
2003-07-18 17:58:37 +00:00
/*
* 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.
*/
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
old = casuword(&umtx->u_owner, owner,
count <= 1 ? UMTX_UNOWNED : UMTX_CONTESTED);
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
umtxq_lock(&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_signal(&key,1);
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
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
- 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
if (old == -1)
return (EFAULT);
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
if (old != owner)
return (EINVAL);
- 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
return (0);
- 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
}
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
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
/*
* Lock a umtx object.
*/
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_umtx32(struct thread *td, uint32_t *m, uint32_t id,
const struct timespec *timeout)
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +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 umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
struct umtx_q *uq;
uint32_t owner;
uint32_t old;
int error = 0;
uq = td->td_umtxq;
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_init(&timo, CLOCK_REALTIME, 0, timeout);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
/*
* Care must be exercised when dealing with umtx structure. It
* can fault on any access.
*/
for (;;) {
/*
* Try the uncontested case. This should be done in userland.
*/
owner = casuword32(m, UMUTEX_UNOWNED, id);
/* The acquire succeeded. */
if (owner == UMUTEX_UNOWNED)
return (0);
/* The address was invalid. */
if (owner == -1)
return (EFAULT);
/* If no one owns it but it is contested try to acquire it. */
if (owner == UMUTEX_CONTESTED) {
owner = casuword32(m,
UMUTEX_CONTESTED, id | UMUTEX_CONTESTED);
if (owner == UMUTEX_CONTESTED)
return (0);
/* The address was invalid. */
if (owner == -1)
return (EFAULT);
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 = umtxq_check_susp(td);
if (error != 0)
break;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
/* If this failed the lock has changed, restart. */
continue;
}
/*
* If we caught a signal, we have retried and now
* exit immediately.
*/
if (error != 0)
return (error);
if ((error = umtx_key_get(m, TYPE_SIMPLE_LOCK,
AUTO_SHARE, &uq->uq_key)) != 0)
return (error);
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_insert(uq);
umtxq_unbusy(&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.
*/
old = casuword32(m, owner, owner | UMUTEX_CONTESTED);
/* The address was invalid. */
if (old == -1) {
umtxq_lock(&uq->uq_key);
umtxq_remove(uq);
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
return (EFAULT);
}
/*
* 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);
if (old == owner)
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, "umtx", timeout == NULL ?
Fix COMPAT_FREEBSD32 build. Submitted by: Andreas Tobler < andreast at fgznet dot ch >
2012-03-30 09:03:53 +00:00
NULL : &timo);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +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)
error = umtxq_check_susp(td);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
}
if (timeout == NULL) {
/* Mutex locking is restarted if it is interrupted. */
if (error == EINTR)
error = ERESTART;
} else {
/* Timed-locking is not restarted. */
if (error == ERESTART)
error = EINTR;
}
return (error);
}
/*
* Unlock a umtx object.
*/
static int
do_unlock_umtx32(struct thread *td, uint32_t *m, uint32_t id)
{
struct umtx_key key;
uint32_t owner;
uint32_t old;
int error;
int count;
/*
* Make sure we own this mtx.
*/
owner = fuword32(m);
if (owner == -1)
return (EFAULT);
if ((owner & ~UMUTEX_CONTESTED) != id)
return (EPERM);
/* This should be done in userland */
if ((owner & UMUTEX_CONTESTED) == 0) {
old = casuword32(m, owner, UMUTEX_UNOWNED);
if (old == -1)
return (EFAULT);
if (old == owner)
return (0);
owner = old;
}
/* We should only ever be in here for contested locks */
if ((error = umtx_key_get(m, TYPE_SIMPLE_LOCK, AUTO_SHARE,
&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.
*/
old = casuword32(m, owner,
count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED);
umtxq_lock(&key);
umtxq_signal(&key,1);
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
if (old == -1)
return (EFAULT);
if (old != owner)
return (EINVAL);
return (0);
}
#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
/*
* 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,
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, 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;
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);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
if (compat32 == 0)
tmp = fuword(addr);
else
Make UMTX_OP_WAIT_UINT actually wait for an unsigned integer on 64-bits machine. MFC after: 1 week
2009-04-13 05:21:17 +00:00
tmp = (unsigned int)fuword32(addr);
Reduce code size by creating common timed sleeping function.
2012-03-29 02:46:43 +00:00
umtxq_lock(&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
if (tmp == id)
error = umtxq_sleep(uq, "uwait", timeout == NULL ?
NULL : &timo);
Reduce code size by creating common timed sleeping function.
2012-03-29 02:46:43 +00:00
if ((uq->uq_flags & UQF_UMTXQ) == 0)
error = 0;
else
umtxq_remove(uq);
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,
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);
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 = 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,
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;
int error = 0;
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;
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
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 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
owner = fuword32(__DEVOLATILE(void *, &m->m_owner));
if (mode == _UMUTEX_WAIT) {
if (owner == UMUTEX_UNOWNED || owner == UMUTEX_CONTESTED)
return (0);
} else {
/*
* Try the uncontested case. This should be done in userland.
*/
owner = casuword32(&m->m_owner, UMUTEX_UNOWNED, id);
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. */
if (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);
/* The address was invalid. */
if (owner == -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 no one owns it but it is contested try to acquire it. */
if (owner == UMUTEX_CONTESTED) {
owner = casuword32(&m->m_owner,
UMUTEX_CONTESTED, id | UMUTEX_CONTESTED);
if (owner == UMUTEX_CONTESTED)
return (0);
/* The address was invalid. */
if (owner == -1)
return (EFAULT);
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 = umtxq_check_susp(td);
if (error != 0)
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
/* If this failed the lock has changed, restart. */
continue;
}
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
if ((flags & UMUTEX_ERROR_CHECK) != 0 &&
(owner & ~UMUTEX_CONTESTED) == id)
return (EDEADLK);
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.
*/
old = casuword32(&m->m_owner, owner, owner | UMUTEX_CONTESTED);
/* The address was invalid. */
if (old == -1) {
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);
return (EFAULT);
}
/*
* 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);
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)
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, "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)
error = umtxq_check_susp(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
}
return (0);
}
/*
* Unlock PTHREAD_PRIO_NONE protocol POSIX mutex.
*/
static int
do_unlock_normal(struct thread *td, struct umutex *m, uint32_t flags)
{
struct umtx_key key;
uint32_t owner, old, id;
int error;
int count;
id = td->td_tid;
/*
* Make sure we own this mtx.
*/
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
owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_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 (owner == -1)
return (EFAULT);
if ((owner & ~UMUTEX_CONTESTED) != id)
return (EPERM);
if ((owner & UMUTEX_CONTESTED) == 0) {
old = casuword32(&m->m_owner, owner, UMUTEX_UNOWNED);
if (old == -1)
return (EFAULT);
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 */
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.
*/
old = casuword32(&m->m_owner, owner,
count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED);
umtxq_lock(&key);
umtxq_signal(&key,1);
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
if (old == -1)
return (EFAULT);
if (old != owner)
return (EINVAL);
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;
owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner));
if (owner == -1)
return (EFAULT);
if ((owner & ~UMUTEX_CONTESTED) != 0)
return (0);
flags = fuword32(&m->m_flags);
/* 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);
if (count <= 1)
owner = casuword32(&m->m_owner, UMUTEX_CONTESTED, UMUTEX_UNOWNED);
umtxq_lock(&key);
if (count != 0 && (owner & ~UMUTEX_CONTESTED) == 0)
umtxq_signal(&key, 1);
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
return (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
/*
* 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;
switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
case 0:
type = TYPE_NORMAL_UMUTEX;
break;
case UMUTEX_PRIO_INHERIT:
type = TYPE_PI_UMUTEX;
break;
case UMUTEX_PRIO_PROTECT:
type = TYPE_PP_UMUTEX;
break;
default:
return (EINVAL);
}
if ((error = umtx_key_get(m, type, GET_SHARE(flags),
&key)) != 0)
return (error);
owner = 0;
umtxq_lock(&key);
umtxq_busy(&key);
count = umtxq_count(&key);
umtxq_unlock(&key);
/*
* 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.
*/
if (count > 1) {
owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner));
while ((owner & UMUTEX_CONTESTED) ==0) {
old = casuword32(&m->m_owner, owner,
owner|UMUTEX_CONTESTED);
if (old == owner)
break;
owner = old;
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 (old == -1)
break;
error = umtxq_check_susp(td);
if (error != 0)
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
}
} else if (count == 1) {
owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner));
while ((owner & ~UMUTEX_CONTESTED) != 0 &&
(owner & UMUTEX_CONTESTED) == 0) {
old = casuword32(&m->m_owner, owner,
owner|UMUTEX_CONTESTED);
if (old == owner)
break;
owner = old;
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 (old == -1)
break;
error = umtxq_check_susp(td);
if (error != 0)
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
}
}
umtxq_lock(&key);
if (owner == -1) {
error = EFAULT;
umtxq_signal(&key, INT_MAX);
}
else if (count != 0 && (owner & ~UMUTEX_CONTESTED) == 0)
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);
}
/*
* 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;
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
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);
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)
panic("pi_ower != NULL");
pi->pi_owner = owner;
TAILQ_INSERT_TAIL(&uq_owner->uq_pi_contested, pi, pi_link);
}
/*
* Claim ownership of a PI mutex.
*/
static int
umtx_pi_claim(struct umtx_pi *pi, struct thread *owner)
{
struct umtx_q *uq, *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_lock_spin(&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) {
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_unlock_spin(&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.
*/
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_unlock_spin(&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) {
int pri;
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
}
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_unlock_spin(&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;
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
mtx_lock_spin(&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);
}
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
mtx_unlock_spin(&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
/*
* Sleep on a PI mutex.
*/
static int
umtxq_sleep_pi(struct umtx_q *uq, struct umtx_pi *pi,
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
uint32_t owner, const char *wmesg, 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 umtxq_chain *uc;
struct thread *td, *td1;
struct umtx_q *uq1;
int pri;
int error = 0;
td = uq->uq_thread;
KASSERT(td == curthread, ("inconsistent uq_thread"));
uc = umtxq_getchain(&uq->uq_key);
UMTXQ_LOCKED_ASSERT(uc);
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_BUSY_ASSERT(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
umtxq_insert(uq);
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
mtx_lock_spin(&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) {
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
mtx_unlock_spin(&umtx_lock);
Create a global thread hash table to speed up thread lookup, use rwlock to protect the table. In old code, thread lookup is done with process lock held, to find a thread, kernel has to iterate through process and thread list, this is quite inefficient. With this change, test shows in extreme case performance is dramatically improved. Earlier patch was reviewed by: jhb, julian
2010-10-09 02:50:23 +00:00
/* XXX Only look up thread in current process. */
td1 = tdfind(owner, curproc->p_pid);
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_lock_spin(&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);
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_unlock_spin(&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);
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_lock_spin(&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);
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_unlock_spin(&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)
{
struct umtxq_chain *uc;
uc = umtxq_getchain(&pi->pi_key);
UMTXQ_LOCKED_ASSERT(uc);
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) {
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_lock_spin(&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) {
TAILQ_REMOVE(&pi->pi_owner->td_umtxq->uq_pi_contested,
pi, pi_link);
pi->pi_owner = NULL;
}
KASSERT(TAILQ_EMPTY(&pi->pi_blocked),
("blocked queue not empty"));
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_unlock_spin(&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;
uint32_t id, owner, old;
int error;
id = td->td_tid;
uq = td->td_umtxq;
if ((error = umtx_key_get(m, TYPE_PI_UMUTEX, GET_SHARE(flags),
&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.
*/
owner = casuword32(&m->m_owner, UMUTEX_UNOWNED, id);
/* The acquire succeeded. */
if (owner == UMUTEX_UNOWNED) {
error = 0;
break;
}
/* The address was invalid. */
if (owner == -1) {
error = EFAULT;
break;
}
/* If no one owns it but it is contested try to acquire it. */
if (owner == UMUTEX_CONTESTED) {
owner = casuword32(&m->m_owner,
UMUTEX_CONTESTED, id | UMUTEX_CONTESTED);
if (owner == UMUTEX_CONTESTED) {
umtxq_lock(&uq->uq_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
umtxq_busy(&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 = umtx_pi_claim(pi, td);
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
umtxq_unlock(&uq->uq_key);
break;
}
/* The address was invalid. */
if (owner == -1) {
error = EFAULT;
break;
}
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 = umtxq_check_susp(td);
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 this failed the lock has changed, restart. */
continue;
}
if ((flags & UMUTEX_ERROR_CHECK) != 0 &&
(owner & ~UMUTEX_CONTESTED) == id) {
error = EDEADLK;
break;
}
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.
*/
old = casuword32(&m->m_owner, owner, owner | UMUTEX_CONTESTED);
/* The address was invalid. */
if (old == -1) {
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
error = EFAULT;
break;
}
umtxq_lock(&uq->uq_key);
/*
* 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.
*/
if (old == owner)
error = umtxq_sleep_pi(uq, pi, owner & ~UMUTEX_CONTESTED,
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
"umtxpi", timeout == NULL ? NULL : &timo);
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
else {
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&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
error = umtxq_check_susp(td);
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
do_unlock_pi(struct thread *td, struct umutex *m, uint32_t flags)
{
struct umtx_key key;
struct umtx_q *uq_first, *uq_first2, *uq_me;
struct umtx_pi *pi, *pi2;
uint32_t owner, old, id;
int error;
int count;
int pri;
id = td->td_tid;
/*
* Make sure we own this mtx.
*/
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
owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_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 (owner == -1)
return (EFAULT);
if ((owner & ~UMUTEX_CONTESTED) != id)
return (EPERM);
/* This should be done in userland */
if ((owner & UMUTEX_CONTESTED) == 0) {
old = casuword32(&m->m_owner, owner, UMUTEX_UNOWNED);
if (old == -1)
return (EFAULT);
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 */
if ((error = umtx_key_get(m, TYPE_PI_UMUTEX, GET_SHARE(flags),
&key)) != 0)
return (error);
umtxq_lock(&key);
umtxq_busy(&key);
count = umtxq_count_pi(&key, &uq_first);
if (uq_first != NULL) {
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
mtx_lock_spin(&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?"));
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 != curthread) {
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
mtx_unlock_spin(&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);
}
uq_me = curthread->td_umtxq;
pi->pi_owner = NULL;
TAILQ_REMOVE(&uq_me->uq_pi_contested, pi, pi_link);
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 &&
(uq_first->uq_flags & UQF_UMTXQ) == 0) {
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);
}
}
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(curthread);
- 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(curthread, 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(curthread);
mtx_unlock_spin(&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);
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.
*/
old = casuword32(&m->m_owner, owner,
count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED);
umtxq_lock(&key);
umtxq_unbusy(&key);
umtxq_unlock(&key);
umtx_key_release(&key);
if (old == -1)
return (EFAULT);
if (old != owner)
return (EINVAL);
return (0);
}
/*
* 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;
int error, pri, old_inherited_pri, su;
id = td->td_tid;
uq = td->td_umtxq;
if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags),
&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);
ceiling = RTP_PRIO_MAX - fuword32(&m->m_ceilings[0]);
if (ceiling > RTP_PRIO_MAX) {
error = EINVAL;
goto out;
}
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_lock_spin(&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) {
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_unlock_spin(&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
}
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_unlock_spin(&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
owner = casuword32(&m->m_owner,
UMUTEX_CONTESTED, id | UMUTEX_CONTESTED);
if (owner == UMUTEX_CONTESTED) {
error = 0;
break;
}
/* The address was invalid. */
if (owner == -1) {
error = EFAULT;
break;
}
if ((flags & UMUTEX_ERROR_CHECK) != 0 &&
(owner & ~UMUTEX_CONTESTED) == id) {
error = EDEADLK;
break;
}
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_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);
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_lock_spin(&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);
mtx_unlock_spin(&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 (error != 0) {
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_lock_spin(&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);
mtx_unlock_spin(&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:
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
umtx_key_release(&uq->uq_key);
return (error);
}
/*
* Unlock a PP mutex.
*/
static int
do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags)
{
struct umtx_key key;
struct umtx_q *uq, *uq2;
struct umtx_pi *pi;
uint32_t owner, id;
uint32_t 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.
*/
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
owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_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 (owner == -1)
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;
}
if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags),
&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.
*/
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
error = suword32(__DEVOLATILE(uint32_t *, &m->m_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
umtxq_lock(&key);
if (error == 0)
umtxq_signal(&key, 1);
umtxq_unbusy(&key);
umtxq_unlock(&key);
if (error == -1)
error = EFAULT;
else {
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_lock_spin(&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);
mtx_unlock_spin(&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,
uint32_t *old_ceiling)
{
struct umtx_q *uq;
uint32_t save_ceiling;
uint32_t owner, id;
uint32_t flags;
int error;
flags = fuword32(&m->m_flags);
if ((flags & UMUTEX_PRIO_PROTECT) == 0)
return (EINVAL);
if (ceiling > RTP_PRIO_MAX)
return (EINVAL);
id = td->td_tid;
uq = td->td_umtxq;
if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags),
&uq->uq_key)) != 0)
return (error);
for (;;) {
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
save_ceiling = fuword32(&m->m_ceilings[0]);
owner = casuword32(&m->m_owner,
UMUTEX_CONTESTED, id | UMUTEX_CONTESTED);
if (owner == UMUTEX_CONTESTED) {
suword32(&m->m_ceilings[0], ceiling);
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
suword32(__DEVOLATILE(uint32_t *, &m->m_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
error = 0;
break;
}
/* The address was invalid. */
if (owner == -1) {
error = EFAULT;
break;
}
if ((owner & ~UMUTEX_CONTESTED) == id) {
suword32(&m->m_ceilings[0], ceiling);
error = 0;
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);
if (error == 0 && old_ceiling != NULL)
suword32(old_ceiling, save_ceiling);
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
flags = fuword32(&m->m_flags);
if (flags == -1)
return (EFAULT);
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
switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
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
do_unlock_umutex(struct thread *td, struct umutex *m)
{
uint32_t flags;
flags = fuword32(&m->m_flags);
if (flags == -1)
return (EFAULT);
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
switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
case 0:
return (do_unlock_normal(td, m, flags));
case UMUTEX_PRIO_INHERIT:
return (do_unlock_pi(td, m, flags));
case UMUTEX_PRIO_PROTECT:
return (do_unlock_pp(td, m, 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
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,
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
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;
uint32_t flags;
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
uint32_t clockid;
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;
flags = fuword32(&cv->c_flags);
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) {
clockid = fuword32(&cv->c_clockid);
if (clockid < CLOCK_REALTIME ||
clockid >= CLOCK_THREAD_CPUTIME_ID) {
/* hmm, only HW clock id will work. */
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
*/
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 (fuword32(__DEVOLATILE(uint32_t *, &cv->c_has_waiters)) == 0)
suword32(__DEVOLATILE(uint32_t *, &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
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
error = do_unlock_umutex(td, 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
Remove trailing semicolon, it is a typo.
2012-03-30 12:57:14 +00:00
if (timeout != NULL)
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_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);
suword32(
__DEVOLATILE(uint32_t *,
&cv->c_has_waiters), 0);
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;
flags = fuword32(&cv->c_flags);
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);
error = suword32(
__DEVOLATILE(uint32_t *, &cv->c_has_waiters), 0);
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;
flags = fuword32(&cv->c_flags);
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);
error = suword32(__DEVOLATILE(uint32_t *, &cv->c_has_waiters), 0);
umtxq_lock(&key);
umtxq_unbusy(&key);
umtxq_unlock(&key);
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
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_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;
int error;
uq = td->td_umtxq;
flags = fuword32(&rwlock->rw_flags);
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 (;;) {
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state));
/* try to lock it */
while (!(state & wrflags)) {
if (__predict_false(URWLOCK_READER_COUNT(state) == URWLOCK_MAX_READERS)) {
umtx_key_release(&uq->uq_key);
return (EAGAIN);
}
oldstate = casuword32(&rwlock->rw_state, state, state + 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
if (oldstate == -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
if (oldstate == state) {
umtx_key_release(&uq->uq_key);
return (0);
}
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 = umtxq_check_susp(td);
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
*/
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_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
/* set read contention bit */
while ((state & wrflags) && !(state & URWLOCK_READ_WAITERS)) {
oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_READ_WAITERS);
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 (oldstate == -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
if (oldstate == state)
goto sleep;
state = oldstate;
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 = umtxq_check_susp(td);
if (error != 0)
break;
}
if (error != 0) {
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
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)) {
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&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
error = umtxq_check_susp(td);
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:
/* contention bit is set, before sleeping, increase read waiter count */
blocked_readers = fuword32(&rwlock->rw_blocked_readers);
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;
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state));
}
/* decrease read waiter count, and may clear read contention bit */
blocked_readers = fuword32(&rwlock->rw_blocked_readers);
suword32(&rwlock->rw_blocked_readers, blocked_readers-1);
if (blocked_readers == 1) {
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state));
for (;;) {
oldstate = casuword32(&rwlock->rw_state, state,
state & ~URWLOCK_READ_WAITERS);
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 (oldstate == -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
if (oldstate == state)
break;
state = oldstate;
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 = umtxq_check_susp(td);
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
}
}
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&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;
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;
uq = td->td_umtxq;
flags = fuword32(&rwlock->rw_flags);
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 (;;) {
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state));
while (!(state & URWLOCK_WRITE_OWNER) && URWLOCK_READER_COUNT(state) == 0) {
oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_WRITE_OWNER);
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 (oldstate == -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
if (oldstate == state) {
umtx_key_release(&uq->uq_key);
return (0);
}
state = oldstate;
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 = umtxq_check_susp(td);
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) {
if (!(state & (URWLOCK_WRITE_OWNER|URWLOCK_WRITE_WAITERS)) &&
blocked_readers != 0) {
umtxq_lock(&uq->uq_key);
umtxq_busy(&uq->uq_key);
umtxq_signal_queue(&uq->uq_key, INT_MAX, UMTX_SHARED_QUEUE);
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
/*
* re-read the state, in case it changed between the try-lock above
* and the check below
*/
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_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
while (((state & URWLOCK_WRITE_OWNER) || URWLOCK_READER_COUNT(state) != 0) &&
(state & URWLOCK_WRITE_WAITERS) == 0) {
oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_WRITE_WAITERS);
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 (oldstate == -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
if (oldstate == state)
goto sleep;
state = oldstate;
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 = umtxq_check_susp(td);
if (error != 0)
break;
}
if (error != 0) {
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
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
}
if (!(state & URWLOCK_WRITE_OWNER) && URWLOCK_READER_COUNT(state) == 0) {
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&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
error = umtxq_check_susp(td);
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:
blocked_writers = fuword32(&rwlock->rw_blocked_writers);
suword32(&rwlock->rw_blocked_writers, blocked_writers+1);
while ((state & URWLOCK_WRITE_OWNER) || URWLOCK_READER_COUNT(state) != 0) {
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;
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state));
}
blocked_writers = fuword32(&rwlock->rw_blocked_writers);
suword32(&rwlock->rw_blocked_writers, blocked_writers-1);
if (blocked_writers == 1) {
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state));
for (;;) {
oldstate = casuword32(&rwlock->rw_state, state,
state & ~URWLOCK_WRITE_WAITERS);
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 (oldstate == -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
if (oldstate == state)
break;
state = oldstate;
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 = umtxq_check_susp(td);
/*
* We are leaving the URWLOCK_WRITE_WAITERS
* behind, but this should not harm the
* correctness.
*/
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
blocked_readers = fuword32(&rwlock->rw_blocked_readers);
} 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
umtxq_lock(&uq->uq_key);
umtxq_unbusy(&uq->uq_key);
umtxq_unlock(&uq->uq_key);
}
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;
int error, q, count;
uq = td->td_umtxq;
flags = fuword32(&rwlock->rw_flags);
error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
if (error != 0)
return (error);
state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state));
if (state & URWLOCK_WRITE_OWNER) {
for (;;) {
oldstate = casuword32(&rwlock->rw_state, state,
state & ~URWLOCK_WRITE_OWNER);
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 (oldstate == -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 (oldstate != state) {
state = oldstate;
if (!(oldstate & URWLOCK_WRITE_OWNER)) {
error = EPERM;
goto out;
}
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 = umtxq_check_susp(td);
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 (;;) {
oldstate = casuword32(&rwlock->rw_state, state,
state - 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
if (oldstate == -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 (oldstate != state) {
state = oldstate;
if (URWLOCK_READER_COUNT(oldstate) == 0) {
error = EPERM;
goto out;
}
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 = umtxq_check_susp(td);
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);
}
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;
uint32_t flags, count;
int error;
uq = td->td_umtxq;
flags = fuword32(&sem->_flags);
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);
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);
Eliminate branch and insert an explicit reader memory barrier to ensure that waiter bit is set before reading semaphore count.
2012-01-16 04:39:10 +00:00
casuword32(__DEVOLATILE(uint32_t *, &sem->_has_waiters), 0, 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
count = fuword32(__DEVOLATILE(uint32_t *, &sem->_count));
if (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);
}
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);
umtx_key_release(&uq->uq_key);
return (error);
}
/*
* Signal a userland condition variable.
*/
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;
flags = fuword32(&sem->_flags);
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) {
umtxq_signal(&key, 1);
/*
* 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);
error = suword32(
__DEVOLATILE(uint32_t *, &sem->_has_waiters), 0);
umtxq_lock(&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
}
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
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_lock(struct thread *td, struct _umtx_lock_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
/* struct umtx *umtx */
{
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 do_lock_umtx(td, uap->umtx, td->td_tid, 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
}
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_unlock(struct thread *td, struct _umtx_unlock_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
/* struct umtx *umtx */
{
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
return do_unlock_umtx(td, uap->umtx, td->td_tid);
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 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
__umtx_op_lock_umtx(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
{
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
struct timespec *ts, 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
int error;
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
/* 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);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
if (error != 0)
return (error);
ts = &timeout;
}
return (do_lock_umtx(td, uap->obj, uap->val, ts));
}
static int
__umtx_op_unlock_umtx(struct thread *td, struct _umtx_op_args *uap)
{
return (do_unlock_umtx(td, uap->obj, uap->val));
}
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
}
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_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
}
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_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
}
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_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)
{
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)
{
int count = uap->val;
void *uaddrs[BATCH_SIZE];
char **upp = (char **)uap->obj;
int tocopy;
int error = 0;
int i, pos = 0;
while (count > 0) {
tocopy = count;
if (tocopy > BATCH_SIZE)
tocopy = BATCH_SIZE;
error = copyin(upp+pos, uaddrs, tocopy * sizeof(char *));
if (error != 0)
break;
for (i = 0; i < tocopy; ++i)
kern_umtx_wake(td, uaddrs[i], INT_MAX, 1);
count -= tocopy;
pos += tocopy;
}
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)
{
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
}
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_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 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 do_lock_umutex(td, uap->obj, NULL, _UMUTEX_TRY);
}
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
}
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_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)
{
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)
{
return do_unlock_umutex(td, uap->obj);
}
static int
__umtx_op_set_ceiling(struct thread *td, struct _umtx_op_args *uap)
{
return do_set_ceiling(td, uap->obj, uap->val, uap->uaddr1);
}
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)
{
return do_cv_signal(td, uap->obj);
}
static int
__umtx_op_cv_broadcast(struct thread *td, struct _umtx_op_args *uap)
{
return do_cv_broadcast(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
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)
{
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
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
}
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)
{
return do_sem_wake(td, uap->obj);
}
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)
{
return do_wake2_umutex(td, uap->obj, uap->val);
}
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);
static _umtx_op_func op_table[] = {
__umtx_op_lock_umtx, /* UMTX_OP_LOCK */
__umtx_op_unlock_umtx, /* UMTX_OP_UNLOCK */
__umtx_op_wait, /* UMTX_OP_WAIT */
__umtx_op_wake, /* UMTX_OP_WAKE */
__umtx_op_trylock_umutex, /* UMTX_OP_MUTEX_TRYLOCK */
__umtx_op_lock_umutex, /* UMTX_OP_MUTEX_LOCK */
__umtx_op_unlock_umutex, /* UMTX_OP_MUTEX_UNLOCK */
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_op_set_ceiling, /* UMTX_OP_SET_CEILING */
__umtx_op_cv_wait, /* UMTX_OP_CV_WAIT*/
__umtx_op_cv_signal, /* UMTX_OP_CV_SIGNAL */
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
__umtx_op_cv_broadcast, /* UMTX_OP_CV_BROADCAST */
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_op_wait_uint, /* UMTX_OP_WAIT_UINT */
__umtx_op_rw_rdlock, /* UMTX_OP_RW_RDLOCK */
__umtx_op_rw_wrlock, /* UMTX_OP_RW_WRLOCK */
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
__umtx_op_rw_unlock, /* UMTX_OP_RW_UNLOCK */
__umtx_op_wait_uint_private, /* UMTX_OP_WAIT_UINT_PRIVATE */
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_op_wake_private, /* UMTX_OP_WAKE_PRIVATE */
__umtx_op_wait_umutex, /* UMTX_OP_UMUTEX_WAIT */
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_op_wake_umutex, /* UMTX_OP_UMUTEX_WAKE */
__umtx_op_sem_wait, /* UMTX_OP_SEM_WAIT */
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
__umtx_op_sem_wake, /* UMTX_OP_SEM_WAKE */
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
__umtx_op_nwake_private, /* UMTX_OP_NWAKE_PRIVATE */
__umtx_op_wake2_umutex /* UMTX_OP_UMUTEX_WAKE2 */
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
{
In order to eliminate a branch, convert opcode to unsigned integer.
2006-10-25 06:38:46 +00:00
if ((unsigned)uap->op < UMTX_OP_MAX)
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
int
freebsd32_umtx_lock(struct thread *td, struct freebsd32_umtx_lock_args *uap)
/* struct umtx *umtx */
{
return (do_lock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid, NULL));
}
int
freebsd32_umtx_unlock(struct thread *td, struct freebsd32_umtx_unlock_args *uap)
/* struct umtx *umtx */
{
return (do_unlock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid));
}
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_lock_umtx_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);
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
if (error != 0)
return (error);
ts = &timeout;
}
return (do_lock_umtx32(td, uap->obj, uap->val, ts));
}
static int
__umtx_op_unlock_umtx_compat32(struct thread *td, struct _umtx_op_args *uap)
{
return (do_unlock_umtx32(td, uap->obj, (uint32_t)uap->val));
}
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
}
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_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 {
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
}
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_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
}
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_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
}
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_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
}
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
}
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)
{
int count = uap->val;
uint32_t uaddrs[BATCH_SIZE];
uint32_t **upp = (uint32_t **)uap->obj;
int tocopy;
int error = 0;
int i, pos = 0;
while (count > 0) {
tocopy = count;
if (tocopy > BATCH_SIZE)
tocopy = BATCH_SIZE;
error = copyin(upp+pos, uaddrs, tocopy * sizeof(uint32_t));
if (error != 0)
break;
for (i = 0; i < tocopy; ++i)
kern_umtx_wake(td, (void *)(intptr_t)uaddrs[i],
INT_MAX, 1);
count -= tocopy;
pos += tocopy;
}
return (error);
}
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
static _umtx_op_func op_table_compat32[] = {
__umtx_op_lock_umtx_compat32, /* UMTX_OP_LOCK */
__umtx_op_unlock_umtx_compat32, /* UMTX_OP_UNLOCK */
__umtx_op_wait_compat32, /* UMTX_OP_WAIT */
__umtx_op_wake, /* UMTX_OP_WAKE */
__umtx_op_trylock_umutex, /* UMTX_OP_MUTEX_LOCK */
Fix umtx command order error for freebsd 32bit.
2006-09-22 14:59:10 +00:00
__umtx_op_lock_umutex_compat32, /* UMTX_OP_MUTEX_TRYLOCK */
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
__umtx_op_unlock_umutex, /* UMTX_OP_MUTEX_UNLOCK */
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_op_set_ceiling, /* UMTX_OP_SET_CEILING */
__umtx_op_cv_wait_compat32, /* UMTX_OP_CV_WAIT*/
__umtx_op_cv_signal, /* UMTX_OP_CV_SIGNAL */
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
__umtx_op_cv_broadcast, /* UMTX_OP_CV_BROADCAST */
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_op_wait_compat32, /* UMTX_OP_WAIT_UINT */
__umtx_op_rw_rdlock_compat32, /* UMTX_OP_RW_RDLOCK */
__umtx_op_rw_wrlock_compat32, /* UMTX_OP_RW_WRLOCK */
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
__umtx_op_rw_unlock, /* UMTX_OP_RW_UNLOCK */
__umtx_op_wait_uint_private_compat32, /* UMTX_OP_WAIT_UINT_PRIVATE */
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_op_wake_private, /* UMTX_OP_WAKE_PRIVATE */
__umtx_op_wait_umutex_compat32, /* UMTX_OP_UMUTEX_WAIT */
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_op_wake_umutex, /* UMTX_OP_UMUTEX_WAKE */
__umtx_op_sem_wait_compat32, /* UMTX_OP_SEM_WAIT */
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
__umtx_op_sem_wake, /* UMTX_OP_SEM_WAKE */
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
__umtx_op_nwake_private32, /* UMTX_OP_NWAKE_PRIVATE */
__umtx_op_wake2_umutex /* UMTX_OP_UMUTEX_WAKE2 */
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
};
int
freebsd32_umtx_op(struct thread *td, struct freebsd32_umtx_op_args *uap)
{
In order to eliminate a branch, convert opcode to unsigned integer.
2006-10-25 06:38:46 +00:00
if ((unsigned)uap->op < UMTX_OP_MAX)
Add umtx support for 32bit process on AMD64 machine.
2006-09-22 00:52:54 +00:00
return (*op_table_compat32[uap->op])(td,
(struct _umtx_op_args *)uap);
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)
{
td->td_umtxq = umtxq_alloc();
td->td_umtxq->uq_thread = td;
}
void
umtx_thread_fini(struct thread *td)
{
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.
*/
static void
umtx_exec_hook(void *arg __unused, struct proc *p __unused,
struct image_params *imgp __unused)
{
umtx_thread_cleanup(curthread);
}
/*
* thread_exit() hook.
*/
void
umtx_thread_exit(struct thread *td)
{
umtx_thread_cleanup(td);
}
/*
* clean up umtx data.
*/
static void
umtx_thread_cleanup(struct thread *td)
{
struct umtx_q *uq;
struct umtx_pi *pi;
if ((uq = td->td_umtxq) == NULL)
return;
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_lock_spin(&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 = PRI_MAX;
while ((pi = TAILQ_FIRST(&uq->uq_pi_contested)) != NULL) {
pi->pi_owner = NULL;
TAILQ_REMOVE(&uq->uq_pi_contested, pi, pi_link);
}
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
mtx_unlock_spin(&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
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_MAX);
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
}
Reference in New Issue Copy Permalink