d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
a8d64eb19c
freebsd-skq/sys/kern/subr_turnstile.c

835 lines
24 KiB
C
Raw Normal View History

Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
/*-
* Copyright (c) 1998 Berkeley Software Design, Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Berkeley Software Design Inc's name may not be used to endorse or
* promote products derived from this software without specific prior
* written permission.
*
* THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``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 BERKELEY SOFTWARE DESIGN INC 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.
*
* from BSDI $Id: mutex_witness.c,v 1.1.2.20 2000/04/27 03:10:27 cp Exp $
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
* and BSDI $Id: synch_machdep.c,v 2.3.2.39 2000/04/27 03:10:25 cp Exp $
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
*/
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Implementation of turnstiles used to hold queue of threads blocked on
* non-sleepable locks. Sleepable locks use condition variables to
* implement their queues. Turnstiles differ from a sleep queue in that
* turnstile queue's are assigned to a lock held by an owning thread. Thus,
* when one thread is enqueued onto a turnstile, it can lend its priority
* to the owning thread.
*
* We wish to avoid bloating locks with an embedded turnstile and we do not
* want to use back-pointers in the locks for the same reason. Thus, we
* use a similar approach to that of Solaris 7 as described in Solaris
* Internals by Jim Mauro and Richard McDougall. Turnstiles are looked up
* in a hash table based on the address of the lock. Each entry in the
* hash table is a linked-lists of turnstiles and is called a turnstile
* chain. Each chain contains a spin mutex that protects all of the
* turnstiles in the chain.
*
* Each time a thread is created, a turnstile is malloc'd and attached to
* that thread. When a thread blocks on a lock, if it is the first thread
* to block, it lends its turnstile to the lock. If the lock already has
* a turnstile, then it gives its turnstile to the lock's turnstile's free
Fix a typo in a comment. Submitted by: das
2003-11-12 14:55:45 +00:00
* list. When a thread is woken up, it takes a turnstile from the free list
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* if there are any other waiters. If it is the only thread blocked on the
* lock, then it reclaims the turnstile associated with the lock and removes
* it from the hash table.
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
*/
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
#include "opt_turnstile_profiling.h"
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
#include <sys/param.h>
Mutex profiling code, conditional on the MUTEX_PROFILING option. Adds the following sysctl variables: debug.mutex.prof.enable enable / disable profiling debug.mutex.prof.acquisitions number of mutex acquisitions recorded debug.mutex.prof.records number of acquisition points recorded debug.mutex.prof.maxrecords max number of acquisition points debug.mutex.prof.rejected number of rejections (due to full table) debug.mutex.prof.hashsize hash size debug.mutex.prof.collisions number of hash collisions debug.mutex.prof.stats profiling statistics The code records four numbers for each acquisition point (identified by source file name and line number): longest time held, total time held, number of non-recursive acquisitions, average time held. The measurements are in clock cycles (as returned by get_cyclecount(9)); this may cause measurements on some SMP systems to be unreliable. This can probably be worked around by replacing get_cyclecount(9) by some incarnation of nanotime(9). This work was derived from initial patches by eivind.
2002-04-02 00:01:49 +00:00
#include <sys/systm.h>
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
#include <sys/kernel.h>
Mutex profiling code, conditional on the MUTEX_PROFILING option. Adds the following sysctl variables: debug.mutex.prof.enable enable / disable profiling debug.mutex.prof.acquisitions number of mutex acquisitions recorded debug.mutex.prof.records number of acquisition points recorded debug.mutex.prof.maxrecords max number of acquisition points debug.mutex.prof.rejected number of rejections (due to full table) debug.mutex.prof.hashsize hash size debug.mutex.prof.collisions number of hash collisions debug.mutex.prof.stats profiling statistics The code records four numbers for each acquisition point (identified by source file name and line number): longest time held, total time held, number of non-recursive acquisitions, average time held. The measurements are in clock cycles (as returned by get_cyclecount(9)); this may cause measurements on some SMP systems to be unreliable. This can probably be worked around by replacing get_cyclecount(9) by some incarnation of nanotime(9). This work was derived from initial patches by eivind.
2002-04-02 00:01:49 +00:00
#include <sys/ktr.h>
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
#include <sys/lock.h>
Undo part of the tangle of having sys/lock.h and sys/mutex.h included in other "system" header files. Also help the deprecation of lockmgr.h by making it a sub-include of sys/lock.h and removing sys/lockmgr.h form kernel .c files. Sort sys/*.h includes where possible in affected files. OK'ed by: bde (with reservations)
2001-05-01 08:13:21 +00:00
#include <sys/malloc.h>
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
#include <sys/mutex.h>
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
#include <sys/proc.h>
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
#include <sys/queue.h>
- Create a new scheduler api that is defined in sys/sched.h - Begin moving scheduler specific functionality into sched_4bsd.c - Replace direct manipulation of scheduler data with hooks provided by the new api. - Remove KSE specific state modifications and single runq assumptions from kern_switch.c Reviewed by: -arch
2002-10-12 05:32:24 +00:00
#include <sys/sched.h>
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
#include <sys/sysctl.h>
#include <sys/turnstile.h>
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Constants for the hash table of turnstile chains. TC_SHIFT is a magic
* number chosen because the sleep queue's use the same value for the
* shift. Basically, we ignore the lower 8 bits of the address.
* TC_TABLESIZE must be a power of two for TC_MASK to work properly.
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
#define TC_TABLESIZE 128 /* Must be power of 2. */
#define TC_MASK (TC_TABLESIZE - 1)
#define TC_SHIFT 8
#define TC_HASH(lock) (((uintptr_t)(lock) >> TC_SHIFT) & TC_MASK)
#define TC_LOOKUP(lock) &turnstile_chains[TC_HASH(lock)]
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* There are three different lists of turnstiles as follows. The list
* connected by ts_link entries is a per-thread list of all the turnstiles
* attached to locks that we own. This is used to fixup our priority when
* a lock is released. The other two lists use the ts_hash entries. The
Clarify and tweak some comments.
2004-02-27 16:14:27 +00:00
* first of these two is the turnstile chain list that a turnstile is on
* when it is attached to a lock. The second list to use ts_hash is the
* free list hung off of a turnstile that is attached to a lock.
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
*
* Each turnstile contains two lists of threads. The ts_blocked list is
* a linked list of threads blocked on the turnstile's lock. The
Fixup a comment.
2004-03-12 19:05:46 +00:00
* ts_pending list is a linked list of threads previously awakened by
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* turnstile_signal() or turnstile_wait() that are waiting to be put on
* the run queue.
*
* Locking key:
* c - turnstile chain lock
* q - td_contested lock
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile {
TAILQ_HEAD(, thread) ts_blocked; /* (c + q) Blocked threads. */
TAILQ_HEAD(, thread) ts_pending; /* (c) Pending threads. */
LIST_ENTRY(turnstile) ts_hash; /* (c) Chain and free list. */
LIST_ENTRY(turnstile) ts_link; /* (q) Contested locks. */
LIST_HEAD(, turnstile) ts_free; /* (c) Free turnstiles. */
struct lock_object *ts_lockobj; /* (c) Lock we reference. */
- Close a race where a thread on another CPU could release a contested lock and empty its turnstile while the blocking threads still pointed to the turnstile. If the thread on the first CPU blocked on a lock owned by one of the threads blocked on the turnstile just woken up, then the first CPU could try to manipulate a bogus thread queue in the turnstile during priority propagation. - Update locking notes for ts_owner and always clear ts_owner, not just under INVARIANTS. Tested by: sam (1)
2003-11-12 23:48:42 +00:00
struct thread *ts_owner; /* (c + q) Who owns the lock. */
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
};
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile_chain {
LIST_HEAD(, turnstile) tc_turnstiles; /* List of turnstiles. */
struct mtx tc_lock; /* Spin lock for this chain. */
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
#ifdef TURNSTILE_PROFILING
u_int tc_depth; /* Length of tc_queues. */
u_int tc_max_depth; /* Max length of tc_queues. */
#endif
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
};
- Place back STR string declarations for lock/unlock strings used for KTR_LOCK tracing in order to avoid duplication. - Insert some tracepoints back into the mutex acq/rel code, thus ensuring that we can trace all lock acq/rel's again. - All CURPROC != NULL checks are MPASS()es (under MUTEX_DEBUG) because they signify a serious mutex corruption. - Change up some KASSERT()s to MPASS()es, and vice-versa, depending on the type of problem we're debugging (INVARIANTS is used here to check that the API is being used properly whereas MUTEX_DEBUG is used to ensure that something general isn't happening that will have bad impact on mutex locks). Reminded by: jhb, jake, asmodai
2001-02-11 02:54:16 +00:00
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
#ifdef TURNSTILE_PROFILING
u_int turnstile_max_depth;
SYSCTL_NODE(_debug, OID_AUTO, turnstile, CTLFLAG_RD, 0, "turnstile profiling");
SYSCTL_NODE(_debug_turnstile, OID_AUTO, chains, CTLFLAG_RD, 0,
"turnstile chain stats");
SYSCTL_UINT(_debug_turnstile, OID_AUTO, max_depth, CTLFLAG_RD,
&turnstile_max_depth, 0, "maxmimum depth achieved of a single chain");
#endif
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
static struct mtx td_contested_lock;
static struct turnstile_chain turnstile_chains[TC_TABLESIZE];
Make a bunch of malloc types static. Found by: src/tools/tools/kernxref
2005-02-10 12:02:37 +00:00
static MALLOC_DEFINE(M_TURNSTILE, "turnstiles", "turnstiles");
- Move the MI mutexes sched_lock and Giant from being declared in the various machdep.c's to being declared in kern_mutex.c. - Add a new function mutex_init() used to perform early initialization needed for mutexes such as setting up thread0's contested lock list and initializing MI mutexes. Change the various MD startup routines to call this function instead of duplicating all the code themselves. Tested on: alpha, i386
2002-04-02 22:19:16 +00:00
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
/*
* Prototypes for non-exported routines.
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
static void init_turnstile0(void *dummy);
Oops, this didn't make it into my submit before I committed: Defer creation of the sysctl tree for the turnstile profiling stats until a SI_SUB_LOCK sysinit. Doing it in init_turnstiles() is too early as it is called before mi_startup().
2004-06-29 03:48:49 +00:00
#ifdef TURNSTILE_PROFILING
static void init_turnstile_profiling(void *arg);
#endif
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
static void propagate_priority(struct thread *td);
static int turnstile_adjust_thread(struct turnstile *ts,
struct thread *td);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
static void turnstile_setowner(struct turnstile *ts, struct thread *owner);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Walks the chain of turnstiles and their owners to propagate the priority
* of the thread being blocked to all the threads holding locks that have to
* release their locks before this thread can run again.
*/
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
static void
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
propagate_priority(struct thread *td)
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile_chain *tc;
struct turnstile *ts;
int pri;
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Fix up priority propagation: - Use a better test for determining when a process is running. - Convert some checks to assertions. - Remove unnecessary tests. - Save the priority before acquiring a mutex rather than in msleep(9).
2000-11-30 00:51:16 +00:00
mtx_assert(&sched_lock, MA_OWNED);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
pri = td->td_priority;
ts = td->td_blocked;
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
for (;;) {
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
td = ts->ts_owner;
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
if (td == NULL) {
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
/*
* This really isn't quite right. Really
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
* ought to bump priority of thread that
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* next acquires the lock.
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
*/
return;
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Part 1 of KSE-III The ability to schedule multiple threads per process (one one cpu) by making ALL system calls optionally asynchronous. to come: ia64 and power-pc patches, patches for gdb, test program (in tools) Reviewed by: Almost everyone who counts (at various times, peter, jhb, matt, alfred, mini, bernd, and a cast of thousands) NOTE: this is still Beta code, and contains lots of debugging stuff. expect slight instability in signals..
2002-06-29 17:26:22 +00:00
MPASS(td->td_proc != NULL);
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
MPASS(td->td_proc->p_magic == P_MAGIC);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* XXX: The owner of a turnstile can be stale if it is the
* first thread to grab a slock of a sx lock. In that case
* it is possible for us to be at SSLEEP or some other
* weird state. We should probably just return if the state
* isn't SRUN or SLOCK.
*/
KASSERT(!TD_IS_SLEEPING(td),
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
("sleeping thread (tid %d) owns a non-sleepable lock",
td->td_tid));
Fix up priority propagation: - Use a better test for determining when a process is running. - Convert some checks to assertions. - Remove unnecessary tests. - Save the priority before acquiring a mutex rather than in msleep(9).
2000-11-30 00:51:16 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* If this thread already has higher priority than the
* thread that is being blocked, we are finished.
*/
if (td->td_priority <= pri)
return;
Fix up priority propagation: - Use a better test for determining when a process is running. - Convert some checks to assertions. - Remove unnecessary tests. - Save the priority before acquiring a mutex rather than in msleep(9).
2000-11-30 00:51:16 +00:00
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
/*
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
* Bump this thread's priority.
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
*/
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
sched_lend_prio(td, pri);
/*
* If lock holder is actually running or on the run queue
* then we are done.
*/
if (TD_IS_RUNNING(td) || TD_ON_RUNQ(td)) {
MPASS(td->td_blocked == NULL);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
return;
}
Implement a unified run queue and adjust priority levels accordingly. - All processes go into the same array of queues, with different scheduling classes using different portions of the array. This allows user processes to have their priorities propogated up into interrupt thread range if need be. - I chose 64 run queues as an arbitrary number that is greater than 32. We used to have 4 separate arrays of 32 queues each, so this may not be optimal. The new run queue code was written with this in mind; changing the number of run queues only requires changing constants in runq.h and adjusting the priority levels. - The new run queue code takes the run queue as a parameter. This is intended to be used to create per-cpu run queues. Implement wrappers for compatibility with the old interface which pass in the global run queue structure. - Group the priority level, user priority, native priority (before propogation) and the scheduling class into a struct priority. - Change any hard coded priority levels that I found to use symbolic constants (TTIPRI and TTOPRI). - Remove the curpriority global variable and use that of curproc. This was used to detect when a process' priority had lowered and it should yield. We now effectively yield on every interrupt. - Activate propogate_priority(). It should now have the desired effect without needing to also propogate the scheduling class. - Temporarily comment out the call to vm_page_zero_idle() in the idle loop. It interfered with propogate_priority() because the idle process needed to do a non-blocking acquire of Giant and then other processes would try to propogate their priority onto it. The idle process should not do anything except idle. vm_page_zero_idle() will return in the form of an idle priority kernel thread which is woken up at apprioriate times by the vm system. - Update struct kinfo_proc to the new priority interface. Deliberately change its size by adjusting the spare fields. It remained the same size, but the layout has changed, so userland processes that use it would parse the data incorrectly. The size constraint should really be changed to an arbitrary version number. Also add a debug.sizeof sysctl node for struct kinfo_proc.
2001-02-12 00:20:08 +00:00
- Add an extra check in priority_propagation() for UP systems to ensure we don't end up back at ourselves which would indicate deadlock. - Add the proc lock to the witness dup_list as we may hold more than one process lock at a time. - Don't assert a mutex is owned in _mtx_unlock_sleep() as that is too late. We do the checks in the macros instead.
2001-03-07 02:45:15 +00:00
#ifndef SMP
/*
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
* For UP, we check to see if td is curthread (this shouldn't
- Add an extra check in priority_propagation() for UP systems to ensure we don't end up back at ourselves which would indicate deadlock. - Add the proc lock to the witness dup_list as we may hold more than one process lock at a time. - Don't assert a mutex is owned in _mtx_unlock_sleep() as that is too late. We do the checks in the macros instead.
2001-03-07 02:45:15 +00:00
* ever happen however as it would mean we are in a deadlock.)
*/
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
KASSERT(td != curthread, ("Deadlock detected"));
- Add an extra check in priority_propagation() for UP systems to ensure we don't end up back at ourselves which would indicate deadlock. - Add the proc lock to the witness dup_list as we may hold more than one process lock at a time. - Don't assert a mutex is owned in _mtx_unlock_sleep() as that is too late. We do the checks in the macros instead.
2001-03-07 02:45:15 +00:00
#endif
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* If we aren't blocked on a lock, we should be.
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
*/
Rename the mutex thread and process states to use a more generic 'LOCK' name instead. (e.g., SLOCK instead of SMTX, TD_ON_LOCK() instead of TD_ON_MUTEX()) Eventually a turnstile abstraction will be added that will be shared with mutexes and other types of locks. SLOCK/TDI_LOCK will be used internally by the turnstile code and will not be specific to mutexes. Making the change now ensures that turnstiles can be dropped in at a later date without affecting the ABI of userland applications.
2002-10-02 20:31:47 +00:00
KASSERT(TD_ON_LOCK(td), (
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
"thread %d(%s):%d holds %s but isn't blocked on a lock\n",
td->td_tid, td->td_proc->p_comm, td->td_state,
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
ts->ts_lockobj->lo_name));
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Pick up the lock that td is blocked on.
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
ts = td->td_blocked;
MPASS(ts != NULL);
tc = TC_LOOKUP(ts->ts_lockobj);
mtx_lock_spin(&tc->tc_lock);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
/* Resort td on the list if needed. */
if (!turnstile_adjust_thread(ts, td)) {
- Close a race where a thread on another CPU could release a contested lock and empty its turnstile while the blocking threads still pointed to the turnstile. If the thread on the first CPU blocked on a lock owned by one of the threads blocked on the turnstile just woken up, then the first CPU could try to manipulate a bogus thread queue in the turnstile during priority propagation. - Update locking notes for ts_owner and always clear ts_owner, not just under INVARIANTS. Tested by: sam (1)
2003-11-12 23:48:42 +00:00
mtx_unlock_spin(&tc->tc_lock);
return;
}
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
mtx_unlock_spin(&tc->tc_lock);
}
}
- Close a race where a thread on another CPU could release a contested lock and empty its turnstile while the blocking threads still pointed to the turnstile. If the thread on the first CPU blocked on a lock owned by one of the threads blocked on the turnstile just woken up, then the first CPU could try to manipulate a bogus thread queue in the turnstile during priority propagation. - Update locking notes for ts_owner and always clear ts_owner, not just under INVARIANTS. Tested by: sam (1)
2003-11-12 23:48:42 +00:00
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
/*
* Adjust the thread's position on a turnstile after its priority has been
* changed.
*/
static int
turnstile_adjust_thread(struct turnstile *ts, struct thread *td)
{
struct turnstile_chain *tc;
struct thread *td1, *td2;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
mtx_assert(&sched_lock, MA_OWNED);
MPASS(TD_ON_LOCK(td));
/*
* This thread may not be blocked on this turnstile anymore
* but instead might already be woken up on another CPU
* that is waiting on sched_lock in turnstile_unpend() to
* finish waking this thread up. We can detect this case
* by checking to see if this thread has been given a
* turnstile by either turnstile_signal() or
* turnstile_broadcast(). In this case, treat the thread as
* if it was already running.
*/
if (td->td_turnstile != NULL)
return (0);
/*
* 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.
*/
tc = TC_LOOKUP(ts->ts_lockobj);
mtx_assert(&tc->tc_lock, MA_OWNED);
td1 = TAILQ_PREV(td, threadqueue, td_lockq);
td2 = TAILQ_NEXT(td, td_lockq);
if ((td1 != NULL && td->td_priority < td1->td_priority) ||
(td2 != NULL && td->td_priority > td2->td_priority)) {
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
/*
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
* Remove thread from blocked chain and determine where
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
* it should be moved to.
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_lock_spin(&td_contested_lock);
TAILQ_REMOVE(&ts->ts_blocked, td, td_lockq);
TAILQ_FOREACH(td1, &ts->ts_blocked, td_lockq) {
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
MPASS(td1->td_proc->p_magic == P_MAGIC);
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
if (td1->td_priority > td->td_priority)
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
break;
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
if (td1 == NULL)
TAILQ_INSERT_TAIL(&ts->ts_blocked, td, td_lockq);
else
TAILQ_INSERT_BEFORE(td1, td, td_lockq);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_unlock_spin(&td_contested_lock);
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
if (td1 == NULL)
CTR3(KTR_LOCK,
"turnstile_adjust_thread: td %d put at tail on [%p] %s",
td->td_tid, ts->ts_lockobj, ts->ts_lockobj->lo_name);
else
CTR4(KTR_LOCK,
"turnstile_adjust_thread: td %d moved before %d on [%p] %s",
td->td_tid, td1->td_tid, ts->ts_lockobj,
ts->ts_lockobj->lo_name);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
}
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
return (1);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
}
Mutex profiling code, conditional on the MUTEX_PROFILING option. Adds the following sysctl variables: debug.mutex.prof.enable enable / disable profiling debug.mutex.prof.acquisitions number of mutex acquisitions recorded debug.mutex.prof.records number of acquisition points recorded debug.mutex.prof.maxrecords max number of acquisition points debug.mutex.prof.rejected number of rejections (due to full table) debug.mutex.prof.hashsize hash size debug.mutex.prof.collisions number of hash collisions debug.mutex.prof.stats profiling statistics The code records four numbers for each acquisition point (identified by source file name and line number): longest time held, total time held, number of non-recursive acquisitions, average time held. The measurements are in clock cycles (as returned by get_cyclecount(9)); this may cause measurements on some SMP systems to be unreliable. This can probably be worked around by replacing get_cyclecount(9) by some incarnation of nanotime(9). This work was derived from initial patches by eivind.
2002-04-02 00:01:49 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Early initialization of turnstiles. This is not done via a SYSINIT()
* since this needs to be initialized very early when mutexes are first
* initialized.
Mutex profiling code, conditional on the MUTEX_PROFILING option. Adds the following sysctl variables: debug.mutex.prof.enable enable / disable profiling debug.mutex.prof.acquisitions number of mutex acquisitions recorded debug.mutex.prof.records number of acquisition points recorded debug.mutex.prof.maxrecords max number of acquisition points debug.mutex.prof.rejected number of rejections (due to full table) debug.mutex.prof.hashsize hash size debug.mutex.prof.collisions number of hash collisions debug.mutex.prof.stats profiling statistics The code records four numbers for each acquisition point (identified by source file name and line number): longest time held, total time held, number of non-recursive acquisitions, average time held. The measurements are in clock cycles (as returned by get_cyclecount(9)); this may cause measurements on some SMP systems to be unreliable. This can probably be worked around by replacing get_cyclecount(9) by some incarnation of nanotime(9). This work was derived from initial patches by eivind.
2002-04-02 00:01:49 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
void
init_turnstiles(void)
Instead of get_cyclecount(9), use nanotime(9) to record acquisition and release times. Measurements are made and stored in nanoseconds but presented in microseconds, which should be sufficient for the locks for which we actually want this (those that are held long and / or often). Also, rename some variables and structure members to unit-agnostic names.
2002-04-02 14:42:01 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
int i;
Instead of get_cyclecount(9), use nanotime(9) to record acquisition and release times. Measurements are made and stored in nanoseconds but presented in microseconds, which should be sufficient for the locks for which we actually want this (those that are held long and / or often). Also, rename some variables and structure members to unit-agnostic names.
2002-04-02 14:42:01 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
for (i = 0; i < TC_TABLESIZE; i++) {
LIST_INIT(&turnstile_chains[i].tc_turnstiles);
mtx_init(&turnstile_chains[i].tc_lock, "turnstile chain",
NULL, MTX_SPIN);
Oops, this didn't make it into my submit before I committed: Defer creation of the sysctl tree for the turnstile profiling stats until a SI_SUB_LOCK sysinit. Doing it in init_turnstiles() is too early as it is called before mi_startup().
2004-06-29 03:48:49 +00:00
}
mtx_init(&td_contested_lock, "td_contested", NULL, MTX_SPIN);
thread0.td_turnstile = NULL;
}
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
#ifdef TURNSTILE_PROFILING
Oops, this didn't make it into my submit before I committed: Defer creation of the sysctl tree for the turnstile profiling stats until a SI_SUB_LOCK sysinit. Doing it in init_turnstiles() is too early as it is called before mi_startup().
2004-06-29 03:48:49 +00:00
static void
init_turnstile_profiling(void *arg)
{
struct sysctl_oid *chain_oid;
char chain_name[10];
int i;
for (i = 0; i < TC_TABLESIZE; i++) {
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
snprintf(chain_name, sizeof(chain_name), "%d", i);
chain_oid = SYSCTL_ADD_NODE(NULL,
SYSCTL_STATIC_CHILDREN(_debug_turnstile_chains), OID_AUTO,
chain_name, CTLFLAG_RD, NULL, "turnstile chain stats");
SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
"depth", CTLFLAG_RD, &turnstile_chains[i].tc_depth, 0,
NULL);
SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
"max_depth", CTLFLAG_RD, &turnstile_chains[i].tc_max_depth,
0, NULL);
Unbreak mutex profiling (at least for me). o Always check for null when dereferencing the filename component. o Implement a try-and-backoff method for allocating memory to dump stats to avoid a spin-lock -> sleep-lock mutex lock order panic with WITNESS. Approved by: des, markm (mentor) Not objected: jhb
2003-02-25 22:28:46 +00:00
}
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
}
Oops, this didn't make it into my submit before I committed: Defer creation of the sysctl tree for the turnstile profiling stats until a SI_SUB_LOCK sysinit. Doing it in init_turnstiles() is too early as it is called before mi_startup().
2004-06-29 03:48:49 +00:00
SYSINIT(turnstile_profiling, SI_SUB_LOCK, SI_ORDER_ANY,
init_turnstile_profiling, NULL);
#endif
Mutex profiling code, conditional on the MUTEX_PROFILING option. Adds the following sysctl variables: debug.mutex.prof.enable enable / disable profiling debug.mutex.prof.acquisitions number of mutex acquisitions recorded debug.mutex.prof.records number of acquisition points recorded debug.mutex.prof.maxrecords max number of acquisition points debug.mutex.prof.rejected number of rejections (due to full table) debug.mutex.prof.hashsize hash size debug.mutex.prof.collisions number of hash collisions debug.mutex.prof.stats profiling statistics The code records four numbers for each acquisition point (identified by source file name and line number): longest time held, total time held, number of non-recursive acquisitions, average time held. The measurements are in clock cycles (as returned by get_cyclecount(9)); this may cause measurements on some SMP systems to be unreliable. This can probably be worked around by replacing get_cyclecount(9) by some incarnation of nanotime(9). This work was derived from initial patches by eivind.
2002-04-02 00:01:49 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
static void
init_turnstile0(void *dummy)
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
thread0.td_turnstile = turnstile_alloc();
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
}
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
SYSINIT(turnstile0, SI_SUB_LOCK, SI_ORDER_ANY, init_turnstile0, NULL);
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
/*
* Update a thread on the turnstile list after it's priority has been changed.
* The old priority is passed in as an argument.
*/
void
turnstile_adjust(struct thread *td, u_char oldpri)
{
struct turnstile_chain *tc;
struct turnstile *ts;
mtx_assert(&sched_lock, MA_OWNED);
MPASS(TD_ON_LOCK(td));
/*
* Pick up the lock that td is blocked on.
*/
ts = td->td_blocked;
MPASS(ts != NULL);
tc = TC_LOOKUP(ts->ts_lockobj);
mtx_lock_spin(&tc->tc_lock);
/* Resort the turnstile on the list. */
if (!turnstile_adjust_thread(ts, td)) {
mtx_unlock_spin(&tc->tc_lock);
return;
}
/*
* If our priority was lowered and we are at the head of the
* turnstile, then propagate our new priority up the chain.
* Note that we currently don't try to revoke lent priorities
* when our priority goes up.
*/
if (td == TAILQ_FIRST(&ts->ts_blocked) && td->td_priority < oldpri) {
mtx_unlock_spin(&tc->tc_lock);
propagate_priority(td);
} else
mtx_unlock_spin(&tc->tc_lock);
}
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Set the owner of the lock this turnstile is attached to.
*/
static void
turnstile_setowner(struct turnstile *ts, struct thread *owner)
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_assert(&td_contested_lock, MA_OWNED);
MPASS(owner->td_proc->p_magic == P_MAGIC);
MPASS(ts->ts_owner == NULL);
ts->ts_owner = owner;
LIST_INSERT_HEAD(&owner->td_contested, ts, ts_link);
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
}
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Malloc a turnstile for a new thread, initialize it and return it.
*/
struct turnstile *
turnstile_alloc(void)
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile *ts;
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
ts = malloc(sizeof(struct turnstile), M_TURNSTILE, M_WAITOK | M_ZERO);
TAILQ_INIT(&ts->ts_blocked);
TAILQ_INIT(&ts->ts_pending);
LIST_INIT(&ts->ts_free);
return (ts);
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
}
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Free a turnstile when a thread is destroyed.
*/
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
void
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
turnstile_free(struct turnstile *ts)
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
MPASS(ts != NULL);
MPASS(TAILQ_EMPTY(&ts->ts_blocked));
MPASS(TAILQ_EMPTY(&ts->ts_pending));
free(ts, M_TURNSTILE);
- Switch from using save/disable/restore_intr to using critical_enter/exit and change the u_int mtx_saveintr member of struct mtx to a critical_t mtx_savecrit. - On the alpha we no longer need a custom _get_spin_lock() macro to avoid an extra PAL call, so remove it. - Partially fix using mutexes with WITNESS in modules. Change all the _mtx_{un,}lock_{spin,}_flags() macros to accept explicit file and line parameters and rename them to use a prefix of two underscores. Inside of kern_mutex.c, generate wrapper functions for _mtx_{un,}lock_{spin,}_flags() (only using a prefix of one underscore) that are called from modules. The macros mtx_{un,}lock_{spin,}_flags() are mapped to the __mtx_* macros inside of the kernel to inline the usual case of mutex operations and map to the internal _mtx_* functions in the module case so that modules will use WITNESS and KTR logging if the kernel is compiled with support for it.
2001-03-28 02:40:47 +00:00
}
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
/*
* Lock the turnstile chain associated with the specified lock.
*/
void
turnstile_lock(struct lock_object *lock)
{
struct turnstile_chain *tc;
tc = TC_LOOKUP(lock);
mtx_lock_spin(&tc->tc_lock);
}
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Look up the turnstile for a lock in the hash table locking the associated
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
* turnstile chain along the way. If no turnstile is found in the hash
* table, NULL is returned.
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile *
turnstile_lookup(struct lock_object *lock)
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile_chain *tc;
struct turnstile *ts;
tc = TC_LOOKUP(lock);
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
mtx_assert(&tc->tc_lock, MA_OWNED);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
LIST_FOREACH(ts, &tc->tc_turnstiles, ts_hash)
if (ts->ts_lockobj == lock)
return (ts);
return (NULL);
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
}
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Unlock the turnstile chain associated with a given lock.
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
*/
void
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
turnstile_release(struct lock_object *lock)
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile_chain *tc;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
tc = TC_LOOKUP(lock);
mtx_unlock_spin(&tc->tc_lock);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
}
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Take ownership of a turnstile and adjust the priority of the new
* owner appropriately.
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
*/
void
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
turnstile_claim(struct lock_object *lock)
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile_chain *tc;
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
struct turnstile *ts;
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct thread *td, *owner;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
tc = TC_LOOKUP(lock);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_assert(&tc->tc_lock, MA_OWNED);
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
ts = turnstile_lookup(lock);
MPASS(ts != NULL);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
owner = curthread;
mtx_lock_spin(&td_contested_lock);
turnstile_setowner(ts, owner);
mtx_unlock_spin(&td_contested_lock);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
td = TAILQ_FIRST(&ts->ts_blocked);
MPASS(td != NULL);
MPASS(td->td_proc->p_magic == P_MAGIC);
mtx_unlock_spin(&tc->tc_lock);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Update the priority of the new owner if needed.
*/
mtx_lock_spin(&sched_lock);
if (td->td_priority < owner->td_priority)
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
sched_lend_prio(owner, td->td_priority);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_unlock_spin(&sched_lock);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
/*
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
* Block the current thread on the turnstile assicated with 'lock'. This
* function will context switch and not return until this thread has been
* woken back up. This function must be called with the appropriate
* turnstile chain locked and will return with it unlocked.
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
*/
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
void
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
turnstile_wait(struct lock_object *lock, struct thread *owner)
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile_chain *tc;
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
struct turnstile *ts;
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
struct thread *td, *td1;
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
td = curthread;
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
tc = TC_LOOKUP(lock);
mtx_assert(&tc->tc_lock, MA_OWNED);
MPASS(td->td_turnstile != NULL);
MPASS(owner != NULL);
MPASS(owner->td_proc->p_magic == P_MAGIC);
Refine the turnstile and sleep queue interfaces just a bit: - Add a new _lock() call to each API that locks the associated chain lock for a lock_object pointer or wait channel. The _lookup() functions now require that the chain lock be locked via _lock() when they are called. - Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup the associated queue structure internally via _lookup() rather than accepting a pointer from the caller. For turnstiles, this means that the actual lookup of the turnstile in the hash table is only done when the thread actually blocks rather than being done on each loop iteration in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup() is no longer used outside of the sleep queue code except to implement an assertion in cv_destroy(). - Change sleepq_broadcast() and sleepq_signal() to require that the chain lock is already required. For condition variables, this lets the cv_broadcast() and cv_signal() functions lock the sleep queue chain lock while testing the waiters count. This means that the waiters count internal to condition variables is no longer protected by the interlock mutex and cv_broadcast() and cv_signal() now no longer require that the interlock be held when they are called. This lets consumers of condition variables drop the lock before waking other threads which can result in fewer context switches. MFC after: 1 month
2004-10-12 18:36:20 +00:00
/* Look up the turnstile associated with the lock 'lock'. */
ts = turnstile_lookup(lock);
/*
* If the lock does not already have a turnstile, use this thread's
* turnstile. Otherwise insert the current thread into the
* turnstile already in use by this lock.
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
if (ts == NULL) {
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
#ifdef TURNSTILE_PROFILING
tc->tc_depth++;
if (tc->tc_depth > tc->tc_max_depth) {
tc->tc_max_depth = tc->tc_depth;
if (tc->tc_max_depth > turnstile_max_depth)
turnstile_max_depth = tc->tc_max_depth;
}
#endif
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
ts = td->td_turnstile;
LIST_INSERT_HEAD(&tc->tc_turnstiles, ts, ts_hash);
KASSERT(TAILQ_EMPTY(&ts->ts_pending),
("thread's turnstile has pending threads"));
KASSERT(TAILQ_EMPTY(&ts->ts_blocked),
("thread's turnstile has a non-empty queue"));
KASSERT(LIST_EMPTY(&ts->ts_free),
("thread's turnstile has a non-empty free list"));
KASSERT(ts->ts_lockobj == NULL, ("stale ts_lockobj pointer"));
ts->ts_lockobj = lock;
mtx_lock_spin(&td_contested_lock);
TAILQ_INSERT_TAIL(&ts->ts_blocked, td, td_lockq);
turnstile_setowner(ts, owner);
mtx_unlock_spin(&td_contested_lock);
} else {
TAILQ_FOREACH(td1, &ts->ts_blocked, td_lockq)
if (td1->td_priority > td->td_priority)
break;
mtx_lock_spin(&td_contested_lock);
if (td1 != NULL)
TAILQ_INSERT_BEFORE(td1, td, td_lockq);
else
TAILQ_INSERT_TAIL(&ts->ts_blocked, td, td_lockq);
mtx_unlock_spin(&td_contested_lock);
MPASS(td->td_turnstile != NULL);
LIST_INSERT_HEAD(&ts->ts_free, td->td_turnstile, ts_hash);
MPASS(owner == ts->ts_owner);
Add code to make default mutexes adaptive if the ADAPTIVE_MUTEXES kernel option is used (not on by default). - In the case of trying to lock a mutex, if the MTX_CONTESTED flag is set, then we can safely read the thread pointer from the mtx_lock member while holding sched_lock. We then examine the thread to see if it is currently executing on another CPU. If it is, then we keep looping instead of blocking. - In the case of trying to unlock a mutex, it is now possible for a mutex to have MTX_CONTESTED set in mtx_lock but to not have any threads actually blocked on it, so we need to handle that case. In that case, we just release the lock as if MTX_CONTESTED was not set and return. - We do not adaptively spin on Giant as Giant is held for long times and it slows SMP systems down to a crawl (it was taking several minutes, like 5-10 or so for my test alpha and sparc64 SMP boxes to boot up when they adaptively spinned on Giant). - We only compile in the code to do this for SMP kernels, it doesn't make sense for UP kernels. Tested on: i386, alpha, sparc64
2002-05-21 20:47:11 +00:00
}
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
td->td_turnstile = NULL;
mtx_unlock_spin(&tc->tc_lock);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_lock_spin(&sched_lock);
/*
* Handle race condition where a thread on another CPU that owns
* lock 'lock' could have woken us in between us dropping the
* turnstile chain lock and acquiring the sched_lock.
*/
if (td->td_flags & TDF_TSNOBLOCK) {
td->td_flags &= ~TDF_TSNOBLOCK;
commit the version I actually tested.. Submitted by: davidxu
2002-09-29 23:23:25 +00:00
mtx_unlock_spin(&sched_lock);
return;
}
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
#ifdef notyet
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* If we're borrowing an interrupted thread's VM context, we
* must clean up before going to sleep.
*/
if (td->td_ithd != NULL) {
struct ithd *it = td->td_ithd;
if (it->it_interrupted) {
if (LOCK_LOG_TEST(lock, 0))
CTR3(KTR_LOCK, "%s: %p interrupted %p",
__func__, it, it->it_interrupted);
intr_thd_fixup(it);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
}
}
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
#endif
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/* Save who we are blocked on and switch. */
td->td_blocked = ts;
td->td_lockname = lock->lo_name;
TD_SET_LOCK(td);
propagate_priority(td);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
if (LOCK_LOG_TEST(lock, 0))
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
CTR4(KTR_LOCK, "%s: td %d blocked on [%p] %s", __func__,
td->td_tid, lock, lock->lo_name);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
- Change mi_switch() and sched_switch() to accept an optional thread to switch to. If a non-NULL thread pointer is passed in, then the CPU will switch to that thread directly rather than calling choosethread() to pick a thread to choose to. - Make sched_switch() aware of idle threads and know to do TD_SET_CAN_RUN() instead of sticking them on the run queue rather than requiring all callers of mi_switch() to know to do this if they can be called from an idlethread. - Move constants for arguments to mi_switch() and thread_single() out of the middle of the function prototypes and up above into their own section.
2004-07-02 19:09:50 +00:00
mi_switch(SW_VOL, NULL);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
if (LOCK_LOG_TEST(lock, 0))
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
CTR4(KTR_LOCK, "%s: td %d free from blocked on [%p] %s",
__func__, td->td_tid, lock, lock->lo_name);
If we have already panic'd then don't bother enforcing mutex asserts as things are pretty much shot already and all panic'ing does is hurt our chances of getting a dump. Inspired by: sheldonh
2001-07-31 17:45:50 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_unlock_spin(&sched_lock);
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Pick the highest priority thread on this turnstile and put it on the
* pending list. This must be called with the turnstile chain locked.
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
int
turnstile_signal(struct turnstile *ts)
{
struct turnstile_chain *tc;
struct thread *td;
int empty;
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
MPASS(ts != NULL);
MPASS(curthread->td_proc->p_magic == P_MAGIC);
MPASS(ts->ts_owner == curthread);
tc = TC_LOOKUP(ts->ts_lockobj);
mtx_assert(&tc->tc_lock, MA_OWNED);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Pick the highest priority thread blocked on this lock and
* move it to the pending list.
*/
td = TAILQ_FIRST(&ts->ts_blocked);
MPASS(td->td_proc->p_magic == P_MAGIC);
mtx_lock_spin(&td_contested_lock);
TAILQ_REMOVE(&ts->ts_blocked, td, td_lockq);
mtx_unlock_spin(&td_contested_lock);
TAILQ_INSERT_TAIL(&ts->ts_pending, td, td_lockq);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Force a commit on kern_mutex.c to explain reason for last commit but while I'm at it also add a comment in mtx_validate() explaining the purpose of the last change. Basically, this fixes booting kernels compiled with MUTEX_DEBUG. What used to happen is before we setidt from init386() [still using BTX idt], we called mtx_init() on several mutex locks, notably Giant and some others. This is a problem for MUTEX_DEBUG because it enables mtx_validate() which calls kernacc(), some of which in turn requires Giant. Fix by calling kernacc() from mtx_validate() only if (!cold).
2001-08-24 23:00:59 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* If the turnstile is now empty, remove it from its chain and
* give it to the about-to-be-woken thread. Otherwise take a
* turnstile from the free list and give it to the thread.
Whitespace cleanup.
2002-10-23 10:26:54 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
empty = TAILQ_EMPTY(&ts->ts_blocked);
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
if (empty) {
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
MPASS(LIST_EMPTY(&ts->ts_free));
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
#ifdef TURNSTILE_PROFILING
tc->tc_depth--;
#endif
} else
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
ts = LIST_FIRST(&ts->ts_free);
Assert that the we never give a thread a NULL turnstile when waking it up.
2003-12-09 21:09:54 +00:00
MPASS(ts != NULL);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
LIST_REMOVE(ts, ts_hash);
td->td_turnstile = ts;
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
return (empty);
}
- Add MTX_SYSINIT and SX_SYSINIT as macro glue for allowing sx and mtx locks to be able to setup a SYSINIT call. This helps in places where a lock is needed to protect some data, but the data is not truly associated with a subsystem that can properly initialize it's lock. The macros use the mtx_sysinit() and sx_sysinit() functions, respectively, as the handler argument to SYSINIT(). Reviewed by: alfred, jhb, smp@
2002-04-02 16:05:43 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Put all blocked threads on the pending list. This must be called with
* the turnstile chain locked.
- Add MTX_SYSINIT and SX_SYSINIT as macro glue for allowing sx and mtx locks to be able to setup a SYSINIT call. This helps in places where a lock is needed to protect some data, but the data is not truly associated with a subsystem that can properly initialize it's lock. The macros use the mtx_sysinit() and sx_sysinit() functions, respectively, as the handler argument to SYSINIT(). Reviewed by: alfred, jhb, smp@
2002-04-02 16:05:43 +00:00
*/
void
Rename turnstile_wakeup() to turnstile_broadcast() to make the naming more consistent with other APIs. sleepq and cv's use signal/broadcast, and msleep uses wakeup_one/wakeup. Prior to this turnstiles were using a signal/wakeup mixture.
2004-04-06 19:07:21 +00:00
turnstile_broadcast(struct turnstile *ts)
- Add MTX_SYSINIT and SX_SYSINIT as macro glue for allowing sx and mtx locks to be able to setup a SYSINIT call. This helps in places where a lock is needed to protect some data, but the data is not truly associated with a subsystem that can properly initialize it's lock. The macros use the mtx_sysinit() and sx_sysinit() functions, respectively, as the handler argument to SYSINIT(). Reviewed by: alfred, jhb, smp@
2002-04-02 16:05:43 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct turnstile_chain *tc;
struct turnstile *ts1;
struct thread *td;
- Add MTX_SYSINIT and SX_SYSINIT as macro glue for allowing sx and mtx locks to be able to setup a SYSINIT call. This helps in places where a lock is needed to protect some data, but the data is not truly associated with a subsystem that can properly initialize it's lock. The macros use the mtx_sysinit() and sx_sysinit() functions, respectively, as the handler argument to SYSINIT(). Reviewed by: alfred, jhb, smp@
2002-04-02 16:05:43 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
MPASS(ts != NULL);
MPASS(curthread->td_proc->p_magic == P_MAGIC);
MPASS(ts->ts_owner == curthread);
tc = TC_LOOKUP(ts->ts_lockobj);
mtx_assert(&tc->tc_lock, MA_OWNED);
/*
* Transfer the blocked list to the pending list.
*/
mtx_lock_spin(&td_contested_lock);
TAILQ_CONCAT(&ts->ts_pending, &ts->ts_blocked, td_lockq);
mtx_unlock_spin(&td_contested_lock);
- Add MTX_SYSINIT and SX_SYSINIT as macro glue for allowing sx and mtx locks to be able to setup a SYSINIT call. This helps in places where a lock is needed to protect some data, but the data is not truly associated with a subsystem that can properly initialize it's lock. The macros use the mtx_sysinit() and sx_sysinit() functions, respectively, as the handler argument to SYSINIT(). Reviewed by: alfred, jhb, smp@
2002-04-02 16:05:43 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Give a turnstile to each thread. The last thread gets
* this turnstile.
*/
TAILQ_FOREACH(td, &ts->ts_pending, td_lockq) {
if (LIST_EMPTY(&ts->ts_free)) {
MPASS(TAILQ_NEXT(td, td_lockq) == NULL);
ts1 = ts;
Add two new kernel options to allow rudimentary profiling of the internal hash tables used in the sleep queue and turnstile code. Each option adds a sysctl tree under debug containing the maximum depth of any bucket in the hash table as well as a separate node for each bucket (or chain) containing the current depth and maximum depth for that bucket.
2004-06-29 02:30:12 +00:00
#ifdef TURNSTILE_PROFILING
tc->tc_depth--;
#endif
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
} else
ts1 = LIST_FIRST(&ts->ts_free);
Assert that the we never give a thread a NULL turnstile when waking it up.
2003-12-09 21:09:54 +00:00
MPASS(ts1 != NULL);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
LIST_REMOVE(ts1, ts_hash);
td->td_turnstile = ts1;
}
}
- Close a race where a thread on another CPU could release a contested lock and empty its turnstile while the blocking threads still pointed to the turnstile. If the thread on the first CPU blocked on a lock owned by one of the threads blocked on the turnstile just woken up, then the first CPU could try to manipulate a bogus thread queue in the turnstile during priority propagation. - Update locking notes for ts_owner and always clear ts_owner, not just under INVARIANTS. Tested by: sam (1)
2003-11-12 23:48:42 +00:00
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Wakeup all threads on the pending list and adjust the priority of the
* current thread appropriately. This must be called with the turnstile
* chain locked.
Whitespace cleanup.
2002-10-23 10:26:54 +00:00
*/
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
void
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
turnstile_unpend(struct turnstile *ts)
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
TAILQ_HEAD( ,thread) pending_threads;
struct turnstile_chain *tc;
struct thread *td;
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
u_char cp, pri;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
MPASS(ts != NULL);
MPASS(ts->ts_owner == curthread);
tc = TC_LOOKUP(ts->ts_lockobj);
mtx_assert(&tc->tc_lock, MA_OWNED);
MPASS(!TAILQ_EMPTY(&ts->ts_pending));
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Move the list of pending threads out of the turnstile and
* into a local variable.
*/
TAILQ_INIT(&pending_threads);
TAILQ_CONCAT(&pending_threads, &ts->ts_pending, td_lockq);
#ifdef INVARIANTS
if (TAILQ_EMPTY(&ts->ts_blocked))
ts->ts_lockobj = NULL;
Split the WITNESS and MUTEX_DEBUG options apart so that WITNESS does not depend on MUTEX_DEBUG. The MUTEX_DEBUG option turns on extra assertions and checks to verify that mutexes themselves are implemented properly. The WITNESS option uses extra checks and diagnostics to verify that other code is using mutexes properly.
2000-12-01 00:10:59 +00:00
#endif
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Remove the turnstile from this thread's list of contested locks
* since this thread doesn't own it anymore. New threads will
* not be blocking on the turnstile until it is claimed by a new
* owner.
*/
mtx_lock_spin(&td_contested_lock);
ts->ts_owner = NULL;
LIST_REMOVE(ts, ts_link);
mtx_unlock_spin(&td_contested_lock);
Add a critical section in turnstile_unpend() from before dropping the turnstile chain lock until after making all the awakened threads runnable. First, this fixes a priority inversion race. Second, this attempts to finish waking up all of the threads waiting on a turnstile before doing a preemption. Reviewed by: Stephan Uphoff (who found the priority inversion race)
2004-10-05 18:00:30 +00:00
critical_enter();
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_unlock_spin(&tc->tc_lock);
/*
* Adjust the priority of curthread based on other contested
* locks it owns. Don't lower the priority below the base
* priority however.
*/
td = curthread;
pri = PRI_MAX;
mtx_lock_spin(&sched_lock);
mtx_lock_spin(&td_contested_lock);
LIST_FOREACH(ts, &td->td_contested, ts_link) {
cp = TAILQ_FIRST(&ts->ts_blocked)->td_priority;
if (cp < pri)
pri = cp;
}
mtx_unlock_spin(&td_contested_lock);
Rework the interface between priority propagation (lending) and the schedulers a bit to ensure more correct handling of priorities and fewer priority inversions: - Add two functions to the sched(9) API to handle priority lending: sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these functions to ask the scheduler to lend a thread a set priority and to tell the scheduler when it thinks it is ok for a thread to stop borrowing priority. The unlend case is slightly complex in that the turnstile code tells the scheduler what the minimum priority of the thread needs to be to satisfy the requirements of any other threads blocked on locks owned by the thread in question. The scheduler then decides where the thread can go back to normal mode (if it's normal priority is high enough to satisfy the pending lock requests) or it it should continue to use the priority specified to the sched_unlend_prio() call. This involves adding a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag for priority elevation. - Schedulers now refuse to lower the priority of a thread that is currently borrowing another therad's priority. - If a scheduler changes the priority of a thread that is currently sitting on a turnstile, it will call a new function turnstile_adjust() to inform the turnstile code of the change. This function resorts the thread on the priority list of the turnstile if needed, and if the thread ends up at the head of the list (due to having the highest priority) and its priority was raised, then it will propagate that new priority to the owner of the lock it is blocked on. Some additional fixes specific to the 4BSD scheduler include: - Common code for updating the priority of a thread when the user priority of its associated kse group has been consolidated in a new static function resetpriority_thread(). One change to this function is that it will now only adjust the priority of a thread if it already has a time sharing priority, thus preserving any boosts from a tsleep() until the thread returns to userland. Also, resetpriority() no longer calls maybe_resched() on each thread in the group. Instead, the code calling resetpriority() is responsible for calling resetpriority_thread() on any threads that need to be updated. - schedcpu() now uses resetpriority_thread() instead of just calling sched_prio() directly after it updates a kse group's user priority. - sched_clock() now uses resetpriority_thread() rather than writing directly to td_priority. - sched_nice() now updates all the priorities of the threads after the group priority has been adjusted. Discussed with: bde Reviewed by: ups, jeffr Tested on: 4bsd, ule Tested on: i386, alpha, sparc64
2004-12-30 20:52:44 +00:00
sched_unlend_prio(td, pri);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
/*
* Wake up all the pending threads. If a thread is not blocked
* on a lock, then it is currently executing on another CPU in
Adjust an assertion for the TDF_TSNOBLOCK race handling in turnstile_unpend(). A racing thread that does not have TDI_LOCK set may either be running on another CPU or it may be sitting on a run queue if it was preempted during the very small window in turnstile_wait() between unlocking the turnstile chain lock and locking sched_lock.
2003-12-09 21:14:31 +00:00
* turnstile_wait() or sitting on a run queue waiting to resume
* in turnstile_wait(). Set a flag to force it to try to acquire
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* the lock again instead of blocking.
*/
while (!TAILQ_EMPTY(&pending_threads)) {
td = TAILQ_FIRST(&pending_threads);
TAILQ_REMOVE(&pending_threads, td, td_lockq);
MPASS(td->td_proc->p_magic == P_MAGIC);
if (TD_ON_LOCK(td)) {
td->td_blocked = NULL;
td->td_lockname = NULL;
TD_CLR_LOCK(td);
MPASS(TD_CAN_RUN(td));
Give setrunqueue() and sched_add() more of a clue as to where they are coming from and what is expected from them. MFC after: 2 days
2004-09-01 02:11:28 +00:00
setrunqueue(td, SRQ_BORING);
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
} else {
td->td_flags |= TDF_TSNOBLOCK;
Adjust an assertion for the TDF_TSNOBLOCK race handling in turnstile_unpend(). A racing thread that does not have TDI_LOCK set may either be running on another CPU or it may be sitting on a run queue if it was preempted during the very small window in turnstile_wait() between unlocking the turnstile chain lock and locking sched_lock.
2003-12-09 21:14:31 +00:00
MPASS(TD_IS_RUNNING(td) || TD_ON_RUNQ(td));
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
}
}
Add a critical section in turnstile_unpend() from before dropping the turnstile chain lock until after making all the awakened threads runnable. First, this fixes a priority inversion race. Second, this attempts to finish waking up all of the threads waiting on a turnstile before doing a preemption. Reviewed by: Stephan Uphoff (who found the priority inversion race)
2004-10-05 18:00:30 +00:00
critical_exit();
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
mtx_unlock_spin(&sched_lock);
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
/*
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
* Return the first thread in a turnstile.
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
*/
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
struct thread *
turnstile_head(struct turnstile *ts)
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
{
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
#ifdef INVARIANTS
struct turnstile_chain *tc;
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
MPASS(ts != NULL);
tc = TC_LOOKUP(ts->ts_lockobj);
mtx_assert(&tc->tc_lock, MA_OWNED);
#endif
return (TAILQ_FIRST(&ts->ts_blocked));
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Reference in New Issue Copy Permalink