2000-09-07 01:33:02 +00:00
|
|
|
/*-
|
2017-11-27 15:20:12 +00:00
|
|
|
* SPDX-License-Identifier: BSD-3-Clause
|
|
|
|
*
|
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 $
|
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.
|
|
|
|
*
|
2007-05-18 06:32:24 +00:00
|
|
|
* Each time a thread is created, a turnstile is allocated from a UMA zone
|
|
|
|
* 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 list. When a thread is woken up, it takes a turnstile from
|
|
|
|
* the free list 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.
|
2000-09-07 01:33:02 +00:00
|
|
|
*/
|
|
|
|
|
2003-06-11 00:56:59 +00:00
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
2006-01-27 22:42:12 +00:00
|
|
|
#include "opt_ddb.h"
|
|
|
|
#include "opt_turnstile_profiling.h"
|
2007-06-12 23:27:31 +00:00
|
|
|
#include "opt_sched.h"
|
2006-01-27 22:42:12 +00:00
|
|
|
|
2000-09-07 01:33:02 +00:00
|
|
|
#include <sys/param.h>
|
2002-04-02 00:01:49 +00:00
|
|
|
#include <sys/systm.h>
|
2012-04-14 11:21:24 +00:00
|
|
|
#include <sys/kdb.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>
|
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>
|
|
|
|
#include <sys/mutex.h>
|
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>
|
2002-10-12 05:32:24 +00:00
|
|
|
#include <sys/sched.h>
|
2012-05-15 01:30:25 +00:00
|
|
|
#include <sys/sdt.h>
|
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
|
|
|
|
2007-05-18 06:32:24 +00:00
|
|
|
#include <vm/uma.h>
|
|
|
|
|
2006-01-27 22:42:12 +00:00
|
|
|
#ifdef DDB
|
|
|
|
#include <ddb/ddb.h>
|
2006-08-15 18:29:01 +00:00
|
|
|
#include <sys/lockmgr.h>
|
|
|
|
#include <sys/sx.h>
|
2006-01-27 22:42:12 +00:00
|
|
|
#endif
|
|
|
|
|
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.
|
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)]
|
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
|
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
|
|
|
*
|
2006-01-27 22:42:12 +00:00
|
|
|
* Each turnstile contains three lists of threads. The two ts_blocked lists
|
|
|
|
* are linked list of threads blocked on the turnstile's lock. One list is
|
|
|
|
* for exclusive waiters, and the other is for shared waiters. The
|
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 {
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
struct mtx ts_lock; /* Spin lock for self. */
|
2006-01-27 22:42:12 +00:00
|
|
|
struct threadqueue ts_blocked[2]; /* (c + q) Blocked threads. */
|
|
|
|
struct threadqueue ts_pending; /* (c) Pending threads. */
|
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_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. */
|
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. */
|
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
|
|
|
};
|
2001-02-11 02:54:16 +00:00
|
|
|
|
2004-06-29 02:30:12 +00:00
|
|
|
#ifdef TURNSTILE_PROFILING
|
|
|
|
u_int turnstile_max_depth;
|
2011-11-07 15:43:11 +00:00
|
|
|
static SYSCTL_NODE(_debug, OID_AUTO, turnstile, CTLFLAG_RD, 0,
|
|
|
|
"turnstile profiling");
|
|
|
|
static SYSCTL_NODE(_debug_turnstile, OID_AUTO, chains, CTLFLAG_RD, 0,
|
2004-06-29 02:30:12 +00:00
|
|
|
"turnstile chain stats");
|
|
|
|
SYSCTL_UINT(_debug_turnstile, OID_AUTO, max_depth, CTLFLAG_RD,
|
2012-04-14 23:59:58 +00:00
|
|
|
&turnstile_max_depth, 0, "maximum depth achieved of a single chain");
|
2004-06-29 02:30:12 +00:00
|
|
|
#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];
|
2007-05-18 06:32:24 +00:00
|
|
|
static uma_zone_t turnstile_zone;
|
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);
|
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);
|
2006-01-27 22:42:12 +00:00
|
|
|
static struct thread *turnstile_first_waiter(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
|
|
|
static void turnstile_setowner(struct turnstile *ts, struct thread *owner);
|
2007-05-18 06:32:24 +00:00
|
|
|
#ifdef INVARIANTS
|
|
|
|
static void turnstile_dtor(void *mem, int size, void *arg);
|
|
|
|
#endif
|
|
|
|
static int turnstile_init(void *mem, int size, int flags);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
static void turnstile_fini(void *mem, int size);
|
- 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
|
|
|
|
2012-05-15 01:30:25 +00:00
|
|
|
SDT_PROVIDER_DECLARE(sched);
|
2013-11-26 08:46:27 +00:00
|
|
|
SDT_PROBE_DEFINE(sched, , , sleep);
|
|
|
|
SDT_PROBE_DEFINE2(sched, , , wakeup, "struct thread *",
|
2012-05-15 01:30:25 +00:00
|
|
|
"struct proc *");
|
|
|
|
|
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
|
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 *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
|
|
|
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
THREAD_LOCK_ASSERT(td, 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;
|
2008-02-07 06:55:38 +00:00
|
|
|
THREAD_LOCKPTR_ASSERT(td, &ts->ts_lock);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
/*
|
|
|
|
* Grab a recursive lock on this turnstile chain so it stays locked
|
|
|
|
* for the whole operation. The caller expects us to return with
|
|
|
|
* the original lock held. We only ever lock down the chain so
|
|
|
|
* the lock order is constant.
|
|
|
|
*/
|
|
|
|
mtx_lock_spin(&ts->ts_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
|
|
|
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
|
|
|
|
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
|
|
|
/*
|
2006-01-27 22:42:12 +00:00
|
|
|
* This might be a read lock with no owner. There's
|
|
|
|
* not much we can do, so just bail.
|
- 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
|
|
|
*/
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_unlock_spin(&ts->ts_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
|
|
|
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_lock_flags(td, MTX_DUPOK);
|
|
|
|
mtx_unlock_spin(&ts->ts_lock);
|
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);
|
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
|
|
|
|
|
|
|
/*
|
2006-03-29 23:24:55 +00:00
|
|
|
* If the thread is asleep, then we are probably about
|
2013-02-17 21:37:32 +00:00
|
|
|
* to deadlock. To make debugging this easier, show
|
|
|
|
* backtrace of misbehaving thread and panic to not
|
|
|
|
* leave the kernel deadlocked.
|
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
|
|
|
*/
|
2006-03-29 23:24:55 +00:00
|
|
|
if (TD_IS_SLEEPING(td)) {
|
|
|
|
printf(
|
|
|
|
"Sleeping thread (tid %d, pid %d) owns a non-sleepable lock\n",
|
|
|
|
td->td_tid, td->td_proc->p_pid);
|
2012-04-12 17:43:59 +00:00
|
|
|
kdb_backtrace_thread(td);
|
2006-03-29 23:24:55 +00:00
|
|
|
panic("sleeping thread");
|
|
|
|
}
|
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.
|
|
|
|
*/
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
if (td->td_priority <= pri) {
|
|
|
|
thread_unlock(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
|
|
|
return;
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
}
|
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);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_unlock(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
|
|
|
return;
|
|
|
|
}
|
2001-02-12 00:20:08 +00:00
|
|
|
|
2001-03-07 02:45:15 +00:00
|
|
|
#ifndef SMP
|
|
|
|
/*
|
2001-09-12 08:38:13 +00:00
|
|
|
* For UP, we check to see if td is curthread (this shouldn't
|
2001-03-07 02:45:15 +00:00
|
|
|
* ever happen however as it would mean we are in a deadlock.)
|
|
|
|
*/
|
2001-09-12 08:38:13 +00:00
|
|
|
KASSERT(td != curthread, ("Deadlock detected"));
|
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
|
|
|
*/
|
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",
|
2007-11-14 06:21:24 +00:00
|
|
|
td->td_tid, td->td_name, 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);
|
2008-02-07 06:55:38 +00:00
|
|
|
THREAD_LOCKPTR_ASSERT(td, &ts->ts_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
|
|
|
/* Resort td on the list if needed. */
|
|
|
|
if (!turnstile_adjust_thread(ts, td)) {
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_unlock_spin(&ts->ts_lock);
|
2003-11-12 23:48:42 +00:00
|
|
|
return;
|
|
|
|
}
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
/* The thread lock is released as ts lock above. */
|
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
|
|
|
}
|
|
|
|
}
|
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 thread *td1, *td2;
|
2006-01-27 22:42:12 +00:00
|
|
|
int queue;
|
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
|
|
|
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
THREAD_LOCK_ASSERT(td, MA_OWNED);
|
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
|
|
|
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
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
* that is waiting on the thread lock in turnstile_unpend() to
|
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
|
|
|
* 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.
|
|
|
|
*/
|
2008-02-07 06:55:38 +00:00
|
|
|
THREAD_LOCKPTR_ASSERT(td, &ts->ts_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
|
|
|
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
|
|
|
|
|
|
|
/*
|
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
|
|
|
*/
|
2006-01-27 22:42:12 +00:00
|
|
|
queue = td->td_tsqueue;
|
|
|
|
MPASS(queue == TS_EXCLUSIVE_QUEUE || queue == TS_SHARED_QUEUE);
|
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);
|
2006-01-27 22:42:12 +00:00
|
|
|
TAILQ_REMOVE(&ts->ts_blocked[queue], td, td_lockq);
|
|
|
|
TAILQ_FOREACH(td1, &ts->ts_blocked[queue], td_lockq) {
|
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)
|
2006-01-27 22:42:12 +00:00
|
|
|
TAILQ_INSERT_TAIL(&ts->ts_blocked[queue], td, td_lockq);
|
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
|
|
|
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
|
|
|
}
|
|
|
|
|
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.
|
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)
|
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;
|
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);
|
2004-06-29 03:48:49 +00:00
|
|
|
}
|
|
|
|
mtx_init(&td_contested_lock, "td_contested", NULL, MTX_SPIN);
|
2006-01-17 16:47:42 +00:00
|
|
|
LIST_INIT(&thread0.td_contested);
|
2004-06-29 03:48:49 +00:00
|
|
|
thread0.td_turnstile = NULL;
|
|
|
|
}
|
|
|
|
|
2004-06-29 02:30:12 +00:00
|
|
|
#ifdef TURNSTILE_PROFILING
|
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++) {
|
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);
|
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
|
|
|
}
|
2004-06-29 03:48:49 +00:00
|
|
|
SYSINIT(turnstile_profiling, SI_SUB_LOCK, SI_ORDER_ANY,
|
|
|
|
init_turnstile_profiling, NULL);
|
|
|
|
#endif
|
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)
|
2001-03-28 02:40:47 +00:00
|
|
|
{
|
|
|
|
|
2007-05-18 06:32:24 +00:00
|
|
|
turnstile_zone = uma_zcreate("TURNSTILE", sizeof(struct turnstile),
|
2008-09-08 21:40:15 +00:00
|
|
|
NULL,
|
2007-05-18 06:32:24 +00:00
|
|
|
#ifdef INVARIANTS
|
2008-09-08 21:40:15 +00:00
|
|
|
turnstile_dtor,
|
2007-05-18 06:32:24 +00:00
|
|
|
#else
|
2008-09-08 21:40:15 +00:00
|
|
|
NULL,
|
2007-05-18 06:32:24 +00:00
|
|
|
#endif
|
2008-09-08 21:40:15 +00:00
|
|
|
turnstile_init, turnstile_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
|
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();
|
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);
|
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 *ts;
|
|
|
|
|
|
|
|
MPASS(TD_ON_LOCK(td));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Pick up the lock that td is blocked on.
|
|
|
|
*/
|
|
|
|
ts = td->td_blocked;
|
|
|
|
MPASS(ts != NULL);
|
2008-02-07 06:55:38 +00:00
|
|
|
THREAD_LOCKPTR_ASSERT(td, &ts->ts_lock);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_lock, MA_OWNED);
|
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 the turnstile on the list. */
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
if (!turnstile_adjust_thread(ts, 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
|
|
|
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.
|
|
|
|
*/
|
2006-01-27 22:42:12 +00:00
|
|
|
MPASS(td->td_tsqueue == TS_EXCLUSIVE_QUEUE ||
|
|
|
|
td->td_tsqueue == TS_SHARED_QUEUE);
|
|
|
|
if (td == TAILQ_FIRST(&ts->ts_blocked[td->td_tsqueue]) &&
|
|
|
|
td->td_priority < oldpri) {
|
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
|
|
|
propagate_priority(td);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +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
|
|
|
}
|
|
|
|
|
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)
|
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(ts->ts_owner == NULL);
|
2006-01-27 22:42:12 +00:00
|
|
|
|
|
|
|
/* A shared lock might not have an owner. */
|
|
|
|
if (owner == NULL)
|
|
|
|
return;
|
|
|
|
|
|
|
|
MPASS(owner->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
|
|
|
ts->ts_owner = owner;
|
|
|
|
LIST_INSERT_HEAD(&owner->td_contested, ts, ts_link);
|
2001-03-28 02:40:47 +00:00
|
|
|
}
|
|
|
|
|
2007-05-18 06:32:24 +00:00
|
|
|
#ifdef INVARIANTS
|
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
|
|
|
/*
|
2007-05-18 06:32:24 +00:00
|
|
|
* UMA zone item deallocator.
|
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
|
|
|
*/
|
2007-05-18 06:32:24 +00:00
|
|
|
static void
|
|
|
|
turnstile_dtor(void *mem, int size, void *arg)
|
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;
|
2001-03-28 02:40:47 +00:00
|
|
|
|
2007-05-18 06:32:24 +00:00
|
|
|
ts = mem;
|
|
|
|
MPASS(TAILQ_EMPTY(&ts->ts_blocked[TS_EXCLUSIVE_QUEUE]));
|
|
|
|
MPASS(TAILQ_EMPTY(&ts->ts_blocked[TS_SHARED_QUEUE]));
|
|
|
|
MPASS(TAILQ_EMPTY(&ts->ts_pending));
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/*
|
|
|
|
* UMA zone item initializer.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
turnstile_init(void *mem, int size, int flags)
|
|
|
|
{
|
|
|
|
struct turnstile *ts;
|
|
|
|
|
|
|
|
bzero(mem, size);
|
|
|
|
ts = mem;
|
2006-01-27 22:42:12 +00:00
|
|
|
TAILQ_INIT(&ts->ts_blocked[TS_EXCLUSIVE_QUEUE]);
|
|
|
|
TAILQ_INIT(&ts->ts_blocked[TS_SHARED_QUEUE]);
|
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_INIT(&ts->ts_pending);
|
|
|
|
LIST_INIT(&ts->ts_free);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_init(&ts->ts_lock, "turnstile lock", NULL, MTX_SPIN | MTX_RECURSE);
|
2007-05-18 06:32:24 +00:00
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
static void
|
|
|
|
turnstile_fini(void *mem, int size)
|
|
|
|
{
|
|
|
|
struct turnstile *ts;
|
|
|
|
|
|
|
|
ts = mem;
|
|
|
|
mtx_destroy(&ts->ts_lock);
|
|
|
|
}
|
|
|
|
|
2007-05-18 06:32:24 +00:00
|
|
|
/*
|
|
|
|
* Get a turnstile for a new thread.
|
|
|
|
*/
|
|
|
|
struct turnstile *
|
|
|
|
turnstile_alloc(void)
|
|
|
|
{
|
|
|
|
|
|
|
|
return (uma_zalloc(turnstile_zone, M_WAITOK));
|
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.
|
|
|
|
*/
|
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)
|
2001-03-28 02:40:47 +00:00
|
|
|
{
|
|
|
|
|
2007-05-18 06:32:24 +00:00
|
|
|
uma_zfree(turnstile_zone, ts);
|
2001-03-28 02:40:47 +00:00
|
|
|
}
|
|
|
|
|
2004-10-12 18:36:20 +00:00
|
|
|
/*
|
|
|
|
* Lock the turnstile chain associated with the specified lock.
|
|
|
|
*/
|
|
|
|
void
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
turnstile_chain_lock(struct lock_object *lock)
|
2004-10-12 18:36:20 +00:00
|
|
|
{
|
|
|
|
struct turnstile_chain *tc;
|
|
|
|
|
|
|
|
tc = TC_LOOKUP(lock);
|
|
|
|
mtx_lock_spin(&tc->tc_lock);
|
|
|
|
}
|
|
|
|
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
struct turnstile *
|
|
|
|
turnstile_trywait(struct lock_object *lock)
|
|
|
|
{
|
|
|
|
struct turnstile_chain *tc;
|
|
|
|
struct turnstile *ts;
|
|
|
|
|
|
|
|
tc = TC_LOOKUP(lock);
|
|
|
|
mtx_lock_spin(&tc->tc_lock);
|
|
|
|
LIST_FOREACH(ts, &tc->tc_turnstiles, ts_hash)
|
|
|
|
if (ts->ts_lockobj == lock) {
|
|
|
|
mtx_lock_spin(&ts->ts_lock);
|
|
|
|
return (ts);
|
|
|
|
}
|
|
|
|
|
|
|
|
ts = curthread->td_turnstile;
|
|
|
|
MPASS(ts != NULL);
|
|
|
|
mtx_lock_spin(&ts->ts_lock);
|
|
|
|
KASSERT(ts->ts_lockobj == NULL, ("stale ts_lockobj pointer"));
|
|
|
|
ts->ts_lockobj = lock;
|
|
|
|
|
|
|
|
return (ts);
|
|
|
|
}
|
|
|
|
|
2018-05-10 17:55:24 +00:00
|
|
|
struct thread *
|
|
|
|
turnstile_lock(struct turnstile *ts, struct lock_object **lockp)
|
|
|
|
{
|
|
|
|
struct turnstile_chain *tc;
|
|
|
|
struct lock_object *lock;
|
|
|
|
|
|
|
|
if ((lock = ts->ts_lockobj) == NULL)
|
|
|
|
return (NULL);
|
|
|
|
tc = TC_LOOKUP(lock);
|
|
|
|
mtx_lock_spin(&tc->tc_lock);
|
|
|
|
mtx_lock_spin(&ts->ts_lock);
|
|
|
|
if (__predict_false(lock != ts->ts_lockobj)) {
|
|
|
|
mtx_unlock_spin(&tc->tc_lock);
|
|
|
|
mtx_unlock_spin(&ts->ts_lock);
|
|
|
|
return (NULL);
|
|
|
|
}
|
|
|
|
*lockp = lock;
|
|
|
|
return (ts->ts_owner);
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
turnstile_unlock(struct turnstile *ts, struct lock_object *lock)
|
|
|
|
{
|
|
|
|
struct turnstile_chain *tc;
|
|
|
|
|
|
|
|
mtx_assert(&ts->ts_lock, MA_OWNED);
|
|
|
|
mtx_unlock_spin(&ts->ts_lock);
|
|
|
|
if (ts == curthread->td_turnstile)
|
|
|
|
ts->ts_lockobj = NULL;
|
|
|
|
tc = TC_LOOKUP(lock);
|
|
|
|
mtx_unlock_spin(&tc->tc_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
turnstile_assert(struct turnstile *ts)
|
|
|
|
{
|
|
|
|
MPASS(ts->ts_lockobj == NULL);
|
|
|
|
}
|
|
|
|
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
void
|
|
|
|
turnstile_cancel(struct turnstile *ts)
|
|
|
|
{
|
|
|
|
struct turnstile_chain *tc;
|
|
|
|
struct lock_object *lock;
|
|
|
|
|
|
|
|
mtx_assert(&ts->ts_lock, MA_OWNED);
|
|
|
|
|
|
|
|
mtx_unlock_spin(&ts->ts_lock);
|
|
|
|
lock = ts->ts_lockobj;
|
|
|
|
if (ts == curthread->td_turnstile)
|
|
|
|
ts->ts_lockobj = NULL;
|
|
|
|
tc = TC_LOOKUP(lock);
|
|
|
|
mtx_unlock_spin(&tc->tc_lock);
|
|
|
|
}
|
|
|
|
|
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
|
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.
|
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)
|
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);
|
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)
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
if (ts->ts_lockobj == lock) {
|
|
|
|
mtx_lock_spin(&ts->ts_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
|
|
|
return (ts);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +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 (NULL);
|
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.
|
2001-01-21 22:34:43 +00:00
|
|
|
*/
|
|
|
|
void
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
turnstile_chain_unlock(struct lock_object *lock)
|
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
|
|
|
}
|
|
|
|
|
2006-01-27 22:42:12 +00:00
|
|
|
/*
|
|
|
|
* Return a pointer to the thread waiting on this turnstile with the
|
|
|
|
* most important priority or NULL if the turnstile has no waiters.
|
|
|
|
*/
|
|
|
|
static struct thread *
|
|
|
|
turnstile_first_waiter(struct turnstile *ts)
|
|
|
|
{
|
|
|
|
struct thread *std, *xtd;
|
|
|
|
|
|
|
|
std = TAILQ_FIRST(&ts->ts_blocked[TS_SHARED_QUEUE]);
|
|
|
|
xtd = TAILQ_FIRST(&ts->ts_blocked[TS_EXCLUSIVE_QUEUE]);
|
|
|
|
if (xtd == NULL || (std != NULL && std->td_priority < xtd->td_priority))
|
|
|
|
return (std);
|
|
|
|
return (xtd);
|
|
|
|
}
|
|
|
|
|
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
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
turnstile_claim(struct turnstile *ts)
|
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 *td, *owner;
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +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
|
|
|
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_lock, MA_OWNED);
|
|
|
|
MPASS(ts != curthread->td_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
|
|
|
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
|
|
|
|
2006-01-27 22:42:12 +00:00
|
|
|
td = turnstile_first_waiter(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
|
|
|
MPASS(td != NULL);
|
|
|
|
MPASS(td->td_proc->p_magic == P_MAGIC);
|
2008-02-07 06:55:38 +00:00
|
|
|
THREAD_LOCKPTR_ASSERT(td, &ts->ts_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.
|
|
|
|
*/
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_lock(owner);
|
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 (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);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_unlock(owner);
|
|
|
|
tc = TC_LOOKUP(ts->ts_lockobj);
|
|
|
|
mtx_unlock_spin(&ts->ts_lock);
|
|
|
|
mtx_unlock_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
|
|
|
}
|
|
|
|
|
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
|
|
|
/*
|
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
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
turnstile_wait(struct turnstile *ts, struct thread *owner, int queue)
|
- 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;
|
2001-09-12 08:38:13 +00:00
|
|
|
struct thread *td, *td1;
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
struct lock_object *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
|
|
|
|
2001-09-12 08:38:13 +00:00
|
|
|
td = curthread;
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_lock, MA_OWNED);
|
2006-01-27 22:42:12 +00:00
|
|
|
if (owner)
|
|
|
|
MPASS(owner->td_proc->p_magic == P_MAGIC);
|
|
|
|
MPASS(queue == TS_SHARED_QUEUE || queue == TS_EXCLUSIVE_QUEUE);
|
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
|
|
|
|
2004-10-12 18:36:20 +00:00
|
|
|
/*
|
|
|
|
* 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.
|
|
|
|
*/
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
tc = TC_LOOKUP(ts->ts_lockobj);
|
2011-02-04 14:16:41 +00:00
|
|
|
mtx_assert(&tc->tc_lock, MA_OWNED);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
if (ts == td->td_turnstile) {
|
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
|
|
|
LIST_INSERT_HEAD(&tc->tc_turnstiles, ts, ts_hash);
|
|
|
|
KASSERT(TAILQ_EMPTY(&ts->ts_pending),
|
|
|
|
("thread's turnstile has pending threads"));
|
2006-01-27 22:42:12 +00:00
|
|
|
KASSERT(TAILQ_EMPTY(&ts->ts_blocked[TS_EXCLUSIVE_QUEUE]),
|
|
|
|
("thread's turnstile has exclusive waiters"));
|
|
|
|
KASSERT(TAILQ_EMPTY(&ts->ts_blocked[TS_SHARED_QUEUE]),
|
|
|
|
("thread's turnstile has shared waiters"));
|
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
|
|
|
KASSERT(LIST_EMPTY(&ts->ts_free),
|
|
|
|
("thread's turnstile has a non-empty free list"));
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
MPASS(ts->ts_lockobj != 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
|
|
|
mtx_lock_spin(&td_contested_lock);
|
2006-01-27 22:42:12 +00:00
|
|
|
TAILQ_INSERT_TAIL(&ts->ts_blocked[queue], 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
|
|
|
turnstile_setowner(ts, owner);
|
|
|
|
mtx_unlock_spin(&td_contested_lock);
|
|
|
|
} else {
|
2006-01-27 22:42:12 +00:00
|
|
|
TAILQ_FOREACH(td1, &ts->ts_blocked[queue], 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
|
|
|
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
|
2006-01-27 22:42:12 +00:00
|
|
|
TAILQ_INSERT_TAIL(&ts->ts_blocked[queue], td, td_lockq);
|
|
|
|
MPASS(owner == ts->ts_owner);
|
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);
|
|
|
|
MPASS(td->td_turnstile != NULL);
|
|
|
|
LIST_INSERT_HEAD(&ts->ts_free, td->td_turnstile, ts_hash);
|
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
|
|
|
}
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_lock(td);
|
|
|
|
thread_lock_set(td, &ts->ts_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
|
|
|
td->td_turnstile = 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
|
|
|
/* Save who we are blocked on and switch. */
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
lock = ts->ts_lockobj;
|
2006-01-27 22:42:12 +00:00
|
|
|
td->td_tsqueue = queue;
|
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_blocked = ts;
|
|
|
|
td->td_lockname = lock->lo_name;
|
2010-01-09 01:46:38 +00:00
|
|
|
td->td_blktick = ticks;
|
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_SET_LOCK(td);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
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
|
|
|
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
|
|
|
|
2012-05-15 01:30:25 +00:00
|
|
|
SDT_PROBE0(sched, , , sleep);
|
|
|
|
|
2008-02-07 06:55:38 +00:00
|
|
|
THREAD_LOCKPTR_ASSERT(td, &ts->ts_lock);
|
2008-04-17 04:20:10 +00:00
|
|
|
mi_switch(SW_VOL | SWT_TURNSTILE, 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);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_unlock(td);
|
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
|
2006-01-27 22:42:12 +00:00
|
|
|
turnstile_signal(struct turnstile *ts, int queue)
|
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
|
|
|
{
|
2018-05-19 05:00:16 +00:00
|
|
|
struct turnstile_chain *tc __unused;
|
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;
|
|
|
|
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);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_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
|
|
|
MPASS(curthread->td_proc->p_magic == P_MAGIC);
|
2008-02-06 01:02:13 +00:00
|
|
|
MPASS(ts->ts_owner == curthread || ts->ts_owner == NULL);
|
2006-01-27 22:42:12 +00:00
|
|
|
MPASS(queue == TS_SHARED_QUEUE || queue == TS_EXCLUSIVE_QUEUE);
|
- 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.
|
|
|
|
*/
|
2006-01-27 22:42:12 +00:00
|
|
|
td = TAILQ_FIRST(&ts->ts_blocked[queue]);
|
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(td->td_proc->p_magic == P_MAGIC);
|
|
|
|
mtx_lock_spin(&td_contested_lock);
|
2006-01-27 22:42:12 +00:00
|
|
|
TAILQ_REMOVE(&ts->ts_blocked[queue], 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);
|
|
|
|
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
|
|
|
|
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.
|
2002-10-23 10:26:54 +00:00
|
|
|
*/
|
2006-01-27 22:42:12 +00:00
|
|
|
empty = TAILQ_EMPTY(&ts->ts_blocked[TS_EXCLUSIVE_QUEUE]) &&
|
|
|
|
TAILQ_EMPTY(&ts->ts_blocked[TS_SHARED_QUEUE]);
|
2004-06-29 02:30:12 +00:00
|
|
|
if (empty) {
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
tc = TC_LOOKUP(ts->ts_lockobj);
|
|
|
|
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
|
|
|
MPASS(LIST_EMPTY(&ts->ts_free));
|
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);
|
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);
|
|
|
|
}
|
|
|
|
|
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.
|
2002-04-02 16:05:43 +00:00
|
|
|
*/
|
|
|
|
void
|
2006-01-27 22:42:12 +00:00
|
|
|
turnstile_broadcast(struct turnstile *ts, int queue)
|
2002-04-02 16:05:43 +00:00
|
|
|
{
|
2018-05-19 05:00:16 +00:00
|
|
|
struct turnstile_chain *tc __unused;
|
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 *ts1;
|
|
|
|
struct thread *td;
|
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);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_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
|
|
|
MPASS(curthread->td_proc->p_magic == P_MAGIC);
|
2008-02-06 01:02:13 +00:00
|
|
|
MPASS(ts->ts_owner == curthread || ts->ts_owner == NULL);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
/*
|
|
|
|
* We must have the chain locked so that we can remove the empty
|
|
|
|
* turnstile from the hash queue.
|
|
|
|
*/
|
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(ts->ts_lockobj);
|
|
|
|
mtx_assert(&tc->tc_lock, MA_OWNED);
|
2006-01-27 22:42:12 +00:00
|
|
|
MPASS(queue == TS_SHARED_QUEUE || queue == TS_EXCLUSIVE_QUEUE);
|
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
|
|
|
|
|
|
|
/*
|
|
|
|
* Transfer the blocked list to the pending list.
|
|
|
|
*/
|
|
|
|
mtx_lock_spin(&td_contested_lock);
|
2006-01-27 22:42:12 +00:00
|
|
|
TAILQ_CONCAT(&ts->ts_pending, &ts->ts_blocked[queue], 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);
|
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
|
2006-01-27 22:42:12 +00:00
|
|
|
* this turnstile if the turnstile is 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
|
|
|
*/
|
|
|
|
TAILQ_FOREACH(td, &ts->ts_pending, td_lockq) {
|
|
|
|
if (LIST_EMPTY(&ts->ts_free)) {
|
|
|
|
MPASS(TAILQ_NEXT(td, td_lockq) == NULL);
|
|
|
|
ts1 = ts;
|
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);
|
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;
|
|
|
|
}
|
|
|
|
}
|
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.
|
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
|
2018-06-02 22:37:53 +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;
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
struct turnstile *nts;
|
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;
|
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);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_lock, MA_OWNED);
|
2008-02-06 01:02:13 +00:00
|
|
|
MPASS(ts->ts_owner == curthread || ts->ts_owner == 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
|
|
|
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
|
2006-01-27 22:42:12 +00:00
|
|
|
if (TAILQ_EMPTY(&ts->ts_blocked[TS_EXCLUSIVE_QUEUE]) &&
|
|
|
|
TAILQ_EMPTY(&ts->ts_blocked[TS_SHARED_QUEUE]))
|
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 = NULL;
|
2000-12-01 00:10:59 +00:00
|
|
|
#endif
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
/*
|
|
|
|
* 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;
|
|
|
|
thread_lock(td);
|
|
|
|
mtx_lock_spin(&td_contested_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
|
|
|
/*
|
|
|
|
* 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
|
2006-01-27 22:42:12 +00:00
|
|
|
* owner. There might not be a current owner if this is a shared
|
|
|
|
* 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
|
|
|
*/
|
2006-01-27 22:42:12 +00:00
|
|
|
if (ts->ts_owner != NULL) {
|
|
|
|
ts->ts_owner = NULL;
|
|
|
|
LIST_REMOVE(ts, ts_link);
|
|
|
|
}
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
LIST_FOREACH(nts, &td->td_contested, ts_link) {
|
|
|
|
cp = turnstile_first_waiter(nts)->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
|
|
|
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);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_unlock(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
|
|
|
/*
|
|
|
|
* Wake up all the pending threads. If a thread is not blocked
|
|
|
|
* on a lock, then it is currently executing on another CPU in
|
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);
|
2012-05-15 01:30:25 +00:00
|
|
|
SDT_PROBE2(sched, , , wakeup, td, td->td_proc);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_lock(td);
|
2008-02-07 06:55:38 +00:00
|
|
|
THREAD_LOCKPTR_ASSERT(td, &ts->ts_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
|
|
|
MPASS(td->td_proc->p_magic == P_MAGIC);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
MPASS(TD_ON_LOCK(td));
|
|
|
|
TD_CLR_LOCK(td);
|
|
|
|
MPASS(TD_CAN_RUN(td));
|
|
|
|
td->td_blocked = NULL;
|
|
|
|
td->td_lockname = NULL;
|
2010-01-09 01:46:38 +00:00
|
|
|
td->td_blktick = 0;
|
2006-01-27 22:42:12 +00:00
|
|
|
#ifdef INVARIANTS
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
td->td_tsqueue = 0xff;
|
2006-01-27 22:42:12 +00:00
|
|
|
#endif
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
sched_add(td, SRQ_BORING);
|
|
|
|
thread_unlock(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
|
|
|
}
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_unlock_spin(&ts->ts_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
|
|
|
}
|
|
|
|
|
2006-04-18 18:16:54 +00:00
|
|
|
/*
|
|
|
|
* Give up ownership of a turnstile. This must be called with the
|
|
|
|
* turnstile chain locked.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
turnstile_disown(struct turnstile *ts)
|
|
|
|
{
|
|
|
|
struct thread *td;
|
|
|
|
u_char cp, pri;
|
|
|
|
|
|
|
|
MPASS(ts != NULL);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_lock, MA_OWNED);
|
2006-04-18 18:16:54 +00:00
|
|
|
MPASS(ts->ts_owner == curthread);
|
|
|
|
MPASS(TAILQ_EMPTY(&ts->ts_pending));
|
|
|
|
MPASS(!TAILQ_EMPTY(&ts->ts_blocked[TS_EXCLUSIVE_QUEUE]) ||
|
|
|
|
!TAILQ_EMPTY(&ts->ts_blocked[TS_SHARED_QUEUE]));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* 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);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* 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;
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_lock(td);
|
|
|
|
mtx_unlock_spin(&ts->ts_lock);
|
2006-04-18 18:16:54 +00:00
|
|
|
mtx_lock_spin(&td_contested_lock);
|
|
|
|
LIST_FOREACH(ts, &td->td_contested, ts_link) {
|
|
|
|
cp = turnstile_first_waiter(ts)->td_priority;
|
|
|
|
if (cp < pri)
|
|
|
|
pri = cp;
|
|
|
|
}
|
|
|
|
mtx_unlock_spin(&td_contested_lock);
|
|
|
|
sched_unlend_prio(td, pri);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
thread_unlock(td);
|
2006-04-18 18:16:54 +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 *
|
2006-01-27 22:42:12 +00:00
|
|
|
turnstile_head(struct turnstile *ts, int queue)
|
- 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
|
- 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);
|
2006-01-27 22:42:12 +00:00
|
|
|
MPASS(queue == TS_SHARED_QUEUE || queue == TS_EXCLUSIVE_QUEUE);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_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
|
|
|
#endif
|
2006-01-27 22:42:12 +00:00
|
|
|
return (TAILQ_FIRST(&ts->ts_blocked[queue]));
|
|
|
|
}
|
|
|
|
|
2006-04-18 18:16:54 +00:00
|
|
|
/*
|
|
|
|
* Returns true if a sub-queue of a turnstile is empty.
|
|
|
|
*/
|
|
|
|
int
|
|
|
|
turnstile_empty(struct turnstile *ts, int queue)
|
|
|
|
{
|
|
|
|
#ifdef INVARIANTS
|
|
|
|
|
|
|
|
MPASS(ts != NULL);
|
|
|
|
MPASS(queue == TS_SHARED_QUEUE || queue == TS_EXCLUSIVE_QUEUE);
|
Commit 3/14 of sched_lock decomposition.
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:51:44 +00:00
|
|
|
mtx_assert(&ts->ts_lock, MA_OWNED);
|
2006-04-18 18:16:54 +00:00
|
|
|
#endif
|
|
|
|
return (TAILQ_EMPTY(&ts->ts_blocked[queue]));
|
|
|
|
}
|
|
|
|
|
2006-01-27 22:42:12 +00:00
|
|
|
#ifdef DDB
|
|
|
|
static void
|
|
|
|
print_thread(struct thread *td, const char *prefix)
|
|
|
|
{
|
|
|
|
|
|
|
|
db_printf("%s%p (tid %d, pid %d, \"%s\")\n", prefix, td, td->td_tid,
|
2016-02-10 21:16:37 +00:00
|
|
|
td->td_proc->p_pid, td->td_name);
|
2006-01-27 22:42:12 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
print_queue(struct threadqueue *queue, const char *header, const char *prefix)
|
|
|
|
{
|
|
|
|
struct thread *td;
|
|
|
|
|
|
|
|
db_printf("%s:\n", header);
|
|
|
|
if (TAILQ_EMPTY(queue)) {
|
|
|
|
db_printf("%sempty\n", prefix);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
TAILQ_FOREACH(td, queue, td_lockq) {
|
|
|
|
print_thread(td, prefix);
|
|
|
|
}
|
2000-09-07 01:33:02 +00:00
|
|
|
}
|
2006-01-27 22:42:12 +00:00
|
|
|
|
|
|
|
DB_SHOW_COMMAND(turnstile, db_show_turnstile)
|
|
|
|
{
|
|
|
|
struct turnstile_chain *tc;
|
|
|
|
struct turnstile *ts;
|
|
|
|
struct lock_object *lock;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
if (!have_addr)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* First, see if there is an active turnstile for the lock indicated
|
|
|
|
* by the address.
|
|
|
|
*/
|
|
|
|
lock = (struct lock_object *)addr;
|
|
|
|
tc = TC_LOOKUP(lock);
|
|
|
|
LIST_FOREACH(ts, &tc->tc_turnstiles, ts_hash)
|
|
|
|
if (ts->ts_lockobj == lock)
|
|
|
|
goto found;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Second, see if there is an active turnstile at the address
|
|
|
|
* indicated.
|
|
|
|
*/
|
|
|
|
for (i = 0; i < TC_TABLESIZE; i++)
|
|
|
|
LIST_FOREACH(ts, &turnstile_chains[i].tc_turnstiles, ts_hash) {
|
|
|
|
if (ts == (struct turnstile *)addr)
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
|
|
|
|
db_printf("Unable to locate a turnstile via %p\n", (void *)addr);
|
|
|
|
return;
|
|
|
|
found:
|
|
|
|
lock = ts->ts_lockobj;
|
|
|
|
db_printf("Lock: %p - (%s) %s\n", lock, LOCK_CLASS(lock)->lc_name,
|
|
|
|
lock->lo_name);
|
|
|
|
if (ts->ts_owner)
|
|
|
|
print_thread(ts->ts_owner, "Lock Owner: ");
|
|
|
|
else
|
|
|
|
db_printf("Lock Owner: none\n");
|
|
|
|
print_queue(&ts->ts_blocked[TS_SHARED_QUEUE], "Shared Waiters", "\t");
|
|
|
|
print_queue(&ts->ts_blocked[TS_EXCLUSIVE_QUEUE], "Exclusive Waiters",
|
|
|
|
"\t");
|
|
|
|
print_queue(&ts->ts_pending, "Pending Threads", "\t");
|
|
|
|
|
|
|
|
}
|
2006-04-25 20:28:17 +00:00
|
|
|
|
2006-08-15 16:44:18 +00:00
|
|
|
/*
|
|
|
|
* Show all the threads a particular thread is waiting on based on
|
2017-08-24 15:12:16 +00:00
|
|
|
* non-spin locks.
|
2006-08-15 16:44:18 +00:00
|
|
|
*/
|
2006-04-25 20:28:17 +00:00
|
|
|
static void
|
2006-08-15 16:44:18 +00:00
|
|
|
print_lockchain(struct thread *td, const char *prefix)
|
2006-04-25 20:28:17 +00:00
|
|
|
{
|
|
|
|
struct lock_object *lock;
|
|
|
|
struct lock_class *class;
|
|
|
|
struct turnstile *ts;
|
2017-08-24 15:12:16 +00:00
|
|
|
struct thread *owner;
|
2006-04-25 20:28:17 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Follow the chain. We keep walking as long as the thread is
|
2017-08-24 15:12:16 +00:00
|
|
|
* blocked on a lock that has an owner.
|
2006-04-25 20:28:17 +00:00
|
|
|
*/
|
2006-07-12 21:25:24 +00:00
|
|
|
while (!db_pager_quit) {
|
2006-04-25 20:28:17 +00:00
|
|
|
db_printf("%sthread %d (pid %d, %s) ", prefix, td->td_tid,
|
2016-02-10 21:16:37 +00:00
|
|
|
td->td_proc->p_pid, td->td_name);
|
2006-04-25 20:28:17 +00:00
|
|
|
switch (td->td_state) {
|
|
|
|
case TDS_INACTIVE:
|
|
|
|
db_printf("is inactive\n");
|
|
|
|
return;
|
|
|
|
case TDS_CAN_RUN:
|
|
|
|
db_printf("can run\n");
|
|
|
|
return;
|
|
|
|
case TDS_RUNQ:
|
|
|
|
db_printf("is on a run queue\n");
|
|
|
|
return;
|
|
|
|
case TDS_RUNNING:
|
|
|
|
db_printf("running on CPU %d\n", td->td_oncpu);
|
|
|
|
return;
|
|
|
|
case TDS_INHIBITED:
|
|
|
|
if (TD_ON_LOCK(td)) {
|
|
|
|
ts = td->td_blocked;
|
|
|
|
lock = ts->ts_lockobj;
|
|
|
|
class = LOCK_CLASS(lock);
|
|
|
|
db_printf("blocked on lock %p (%s) \"%s\"\n",
|
|
|
|
lock, class->lc_name, lock->lo_name);
|
|
|
|
if (ts->ts_owner == NULL)
|
|
|
|
return;
|
|
|
|
td = ts->ts_owner;
|
|
|
|
break;
|
2017-08-24 15:12:16 +00:00
|
|
|
} else if (TD_ON_SLEEPQ(td)) {
|
|
|
|
if (!lockmgr_chain(td, &owner) &&
|
|
|
|
!sx_chain(td, &owner)) {
|
|
|
|
db_printf("sleeping on %p \"%s\"\n",
|
|
|
|
td->td_wchan, td->td_wmesg);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
if (owner == NULL)
|
|
|
|
return;
|
|
|
|
td = owner;
|
|
|
|
break;
|
2006-04-25 20:28:17 +00:00
|
|
|
}
|
|
|
|
db_printf("inhibited\n");
|
|
|
|
return;
|
|
|
|
default:
|
|
|
|
db_printf("??? (%#x)\n", td->td_state);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-08-15 16:44:18 +00:00
|
|
|
DB_SHOW_COMMAND(lockchain, db_show_lockchain)
|
2006-04-25 20:28:17 +00:00
|
|
|
{
|
|
|
|
struct thread *td;
|
|
|
|
|
|
|
|
/* Figure out which thread to start with. */
|
|
|
|
if (have_addr)
|
2015-05-21 15:16:18 +00:00
|
|
|
td = db_lookup_thread(addr, true);
|
2006-04-25 20:28:17 +00:00
|
|
|
else
|
|
|
|
td = kdb_thread;
|
|
|
|
|
2006-08-15 16:44:18 +00:00
|
|
|
print_lockchain(td, "");
|
2006-04-25 20:28:17 +00:00
|
|
|
}
|
2017-08-24 15:12:16 +00:00
|
|
|
DB_SHOW_ALIAS(sleepchain, db_show_lockchain);
|
2006-04-25 20:28:17 +00:00
|
|
|
|
2008-09-15 22:45:14 +00:00
|
|
|
DB_SHOW_ALL_COMMAND(chains, db_show_allchains)
|
2006-04-25 20:28:17 +00:00
|
|
|
{
|
|
|
|
struct thread *td;
|
|
|
|
struct proc *p;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
i = 1;
|
2007-01-17 14:58:53 +00:00
|
|
|
FOREACH_PROC_IN_SYSTEM(p) {
|
2006-04-25 20:28:17 +00:00
|
|
|
FOREACH_THREAD_IN_PROC(p, td) {
|
2017-08-24 15:12:16 +00:00
|
|
|
if ((TD_ON_LOCK(td) && LIST_EMPTY(&td->td_contested))
|
|
|
|
|| (TD_IS_INHIBITED(td) && TD_ON_SLEEPQ(td))) {
|
2006-04-25 20:28:17 +00:00
|
|
|
db_printf("chain %d:\n", i++);
|
2006-08-15 16:44:18 +00:00
|
|
|
print_lockchain(td, " ");
|
2006-04-25 20:28:17 +00:00
|
|
|
}
|
2006-07-12 21:25:24 +00:00
|
|
|
if (db_pager_quit)
|
|
|
|
return;
|
2006-04-25 20:28:17 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2008-09-15 22:45:14 +00:00
|
|
|
DB_SHOW_ALIAS(allchains, db_show_allchains)
|
2006-04-25 20:28:17 +00:00
|
|
|
|
|
|
|
static void print_waiters(struct turnstile *ts, int indent);
|
|
|
|
|
|
|
|
static void
|
|
|
|
print_waiter(struct thread *td, int indent)
|
|
|
|
{
|
|
|
|
struct turnstile *ts;
|
|
|
|
int i;
|
|
|
|
|
2006-07-12 21:25:24 +00:00
|
|
|
if (db_pager_quit)
|
|
|
|
return;
|
2006-04-25 20:28:17 +00:00
|
|
|
for (i = 0; i < indent; i++)
|
|
|
|
db_printf(" ");
|
|
|
|
print_thread(td, "thread ");
|
|
|
|
LIST_FOREACH(ts, &td->td_contested, ts_link)
|
|
|
|
print_waiters(ts, indent + 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
print_waiters(struct turnstile *ts, int indent)
|
|
|
|
{
|
|
|
|
struct lock_object *lock;
|
|
|
|
struct lock_class *class;
|
|
|
|
struct thread *td;
|
|
|
|
int i;
|
|
|
|
|
2006-07-12 21:25:24 +00:00
|
|
|
if (db_pager_quit)
|
|
|
|
return;
|
2006-04-25 20:28:17 +00:00
|
|
|
lock = ts->ts_lockobj;
|
|
|
|
class = LOCK_CLASS(lock);
|
|
|
|
for (i = 0; i < indent; i++)
|
|
|
|
db_printf(" ");
|
|
|
|
db_printf("lock %p (%s) \"%s\"\n", lock, class->lc_name, lock->lo_name);
|
|
|
|
TAILQ_FOREACH(td, &ts->ts_blocked[TS_EXCLUSIVE_QUEUE], td_lockq)
|
|
|
|
print_waiter(td, indent + 1);
|
|
|
|
TAILQ_FOREACH(td, &ts->ts_blocked[TS_SHARED_QUEUE], td_lockq)
|
|
|
|
print_waiter(td, indent + 1);
|
|
|
|
TAILQ_FOREACH(td, &ts->ts_pending, td_lockq)
|
|
|
|
print_waiter(td, indent + 1);
|
|
|
|
}
|
|
|
|
|
2006-08-15 16:44:18 +00:00
|
|
|
DB_SHOW_COMMAND(locktree, db_show_locktree)
|
2006-04-25 20:28:17 +00:00
|
|
|
{
|
|
|
|
struct lock_object *lock;
|
|
|
|
struct lock_class *class;
|
|
|
|
struct turnstile_chain *tc;
|
|
|
|
struct turnstile *ts;
|
|
|
|
|
|
|
|
if (!have_addr)
|
|
|
|
return;
|
|
|
|
lock = (struct lock_object *)addr;
|
|
|
|
tc = TC_LOOKUP(lock);
|
|
|
|
LIST_FOREACH(ts, &tc->tc_turnstiles, ts_hash)
|
|
|
|
if (ts->ts_lockobj == lock)
|
|
|
|
break;
|
|
|
|
if (ts == NULL) {
|
|
|
|
class = LOCK_CLASS(lock);
|
|
|
|
db_printf("lock %p (%s) \"%s\"\n", lock, class->lc_name,
|
|
|
|
lock->lo_name);
|
|
|
|
} else
|
|
|
|
print_waiters(ts, 0);
|
|
|
|
}
|
2006-01-27 22:42:12 +00:00
|
|
|
#endif
|