d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
8252626fb4
freebsd-skq/sys/kern/kern_lock.c

1482 lines
39 KiB
C
Raw Normal View History

/* -> /*- for copyright notices, minor format tweaks as necessary
2005-01-06 23:35:40 +00:00
/*-
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* Copyright (c) 2008 Attilio Rao <attilio@FreeBSD.org>
* All rights reserved.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* notice(s), this list of conditions and the following disclaimer as
* the first lines of this file unmodified other than the possible
* addition of one or more copyright notices.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
* 2. Redistributions in binary form must reproduce the above copyright
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* notice(s), this list of conditions and the following disclaimer in the
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
* documentation and/or other materials provided with the distribution.
*
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) 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
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
* DAMAGE.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*/
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#include "opt_adaptive_lockmgrs.h"
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#include "opt_ddb.h"
Add software PMC support. New kernel events can be added at various location for sampling or counting. This will for example allow easy system profiling whatever the processor is with known tools like pmcstat(8). Simultaneous usage of software PMC and hardware PMC is possible, for example looking at the lock acquire failure, page fault while sampling on instructions. Sponsored by: NETASQ MFC after: 1 month
2012-03-28 20:58:30 +00:00
#include "opt_hwpmc_hooks.h"
Add the OpenSolaris dtrace lockstat provider. The lockstat provider adds probes for mutexes, reader/writer and shared/exclusive locks to gather contention statistics and other locking information for dtrace scripts, the lockstat(1M) command and other potential consumers. Reviewed by: attilio jhb jb Approved by: gnn (mentor)
2009-05-26 20:28:22 +00:00
#include "opt_kdtrace.h"
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
#include <sys/param.h>
Add missing includes of sys/ktr.h.
2001-10-11 17:53:43 +00:00
#include <sys/ktr.h>
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
#include <sys/lock.h>
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#include <sys/lock_profile.h>
Include <sys/lockmgr.h> for the definitions of the locking interfaces that are implemented here instead of depending on namespace pollution in <sys/lock.h>. Fixed nearby include messes (1 disordered include and 1 unused include).
2002-08-27 09:59:47 +00:00
#include <sys/lockmgr.h>
- machine/mutex.h -> sys/mutex.h - The initial lock_mtx mutex used in the lockmgr code is initialized very early, so use MUTEX_DECLARE() and MTX_COLD.
2000-10-20 07:28:00 +00:00
#include <sys/mutex.h>
Include <sys/lockmgr.h> for the definitions of the locking interfaces that are implemented here instead of depending on namespace pollution in <sys/lock.h>. Fixed nearby include messes (1 disordered include and 1 unused include).
2002-08-27 09:59:47 +00:00
#include <sys/proc.h>
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#include <sys/sleepqueue.h>
- Replace the series of DEBUG_LOCKS hacks which tried to save the vn_lock caller by saving the stack of the last locker/unlocker in lockmgr. We also put the stack in KTR at the moment. Contributed by: Antoine Brodin <antoine.brodin@laposte.net>
2005-08-03 04:48:22 +00:00
#ifdef DEBUG_LOCKS
#include <sys/stack.h>
#endif
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#include <sys/sysctl.h>
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#include <sys/systm.h>
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#include <machine/cpu.h>
Add WITNESS support to lockmgr locking primitive. This support tries to be as parallel as possible with other locking primitives, but there are differences; more specifically: - The base witness support is alredy equipped for allowing lock duplication acquisition as lockmgr rely on this. - In the case of lockmgr_disown() the lock result unlocked by witness even if it is still held by the "kernel context" - In the case of upgrading we can have 3 different situations: * Total unlocking of the shared lock and nothing else * Real witness upgrade if the owner is the first upgrader * Shared unlocking and exclusive locking if the owner is not the first upgrade but it is still allowed to upgrade - LK_DRAIN is basically handled like an exclusive acquisition Additively new options LK_NODUP and LK_NOWITNESS can now be used with lockinit(): LK_NOWITNESS disables WITNESS for the specified lock while LK_NODUP enable duplicated locks tracking. This will require manpages update and a __FreeBSD_version bumping (addressed by further commits). This patch also fixes a problem occurring if a lockmgr is held in exclusive mode and the same owner try to acquire it in shared mode: currently there is a spourious shared locking acquisition while what we really want is a lock downgrade. Probabilly, this situation can be better served with a EDEADLK failing errno return. Side note: first testing on this patch alredy reveleated several LORs reported, so please expect LORs cascades until resolved. NTFS also is reported broken by WITNESS introduction. BTW, NTFS is exposing a lock leak which needs to be fixed, and this patch can help it out if rightly tweaked. Tested by: kris, yar, Scot Hetzel <swhetzel at gmail dot com>
2008-02-06 00:37:14 +00:00
Add a 'show lockmgr' command that dumps the relevant details of a lockmgr lock.
2006-08-15 16:42:16 +00:00
#ifdef DDB
#include <ddb/ddb.h>
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
Add software PMC support. New kernel events can be added at various location for sampling or counting. This will for example allow easy system profiling whatever the processor is with known tools like pmcstat(8). Simultaneous usage of software PMC and hardware PMC is possible, for example looking at the lock acquire failure, page fault while sampling on instructions. Sponsored by: NETASQ MFC after: 1 month
2012-03-28 20:58:30 +00:00
#ifdef HWPMC_HOOKS
#include <sys/pmckern.h>
PMC_SOFT_DECLARE( , , lock, failed);
#endif
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
CTASSERT(((LK_ADAPTIVE | LK_NOSHARE) & LO_CLASSFLAGS) ==
(LK_ADAPTIVE | LK_NOSHARE));
CTASSERT(LK_UNLOCKED == (LK_UNLOCKED &
~(LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS)));
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#define SQ_EXCLUSIVE_QUEUE 0
#define SQ_SHARED_QUEUE 1
#ifndef INVARIANTS
#define _lockmgr_assert(lk, what, file, line)
#define TD_LOCKS_INC(td)
#define TD_LOCKS_DEC(td)
#else
#define TD_LOCKS_INC(td) ((td)->td_locks++)
#define TD_LOCKS_DEC(td) ((td)->td_locks--)
#endif
#define TD_SLOCKS_INC(td) ((td)->td_lk_slocks++)
#define TD_SLOCKS_DEC(td) ((td)->td_lk_slocks--)
#ifndef DEBUG_LOCKS
#define STACK_PRINT(lk)
#define STACK_SAVE(lk)
#define STACK_ZERO(lk)
#else
#define STACK_PRINT(lk) stack_print_ddb(&(lk)->lk_stack)
#define STACK_SAVE(lk) stack_save(&(lk)->lk_stack)
#define STACK_ZERO(lk) stack_zero(&(lk)->lk_stack)
#endif
#define LOCK_LOG2(lk, string, arg1, arg2) \
if (LOCK_LOG_TEST(&(lk)->lock_object, 0)) \
CTR2(KTR_LOCK, (string), (arg1), (arg2))
#define LOCK_LOG3(lk, string, arg1, arg2, arg3) \
if (LOCK_LOG_TEST(&(lk)->lock_object, 0)) \
CTR3(KTR_LOCK, (string), (arg1), (arg2), (arg3))
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
#define GIANT_DECLARE \
int _i = 0; \
WITNESS_SAVE_DECL(Giant)
#define GIANT_RESTORE() do { \
if (_i > 0) { \
while (_i--) \
mtx_lock(&Giant); \
WITNESS_RESTORE(&Giant.lock_object, Giant); \
} \
} while (0)
#define GIANT_SAVE() do { \
if (mtx_owned(&Giant)) { \
WITNESS_SAVE(&Giant.lock_object, Giant); \
while (mtx_owned(&Giant)) { \
_i++; \
mtx_unlock(&Giant); \
} \
} \
} while (0)
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#define LK_CAN_SHARE(x) \
(((x) & LK_SHARE) && (((x) & LK_EXCLUSIVE_WAITERS) == 0 || \
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
((x) & LK_EXCLUSIVE_SPINNERS) == 0 || \
- Use a different encoding for lockmgr options: make them encoded by bit in order to allow per-bit checks on the options flag, in particular in the consumers code [1] - Re-enable the check against TDP_DEADLKTREAT as the anti-waiters starvation patch allows exclusive waiters to override new shared requests. [1] Requested by: pjd, jeff
2008-04-07 14:46:38 +00:00
curthread->td_lk_slocks || (curthread->td_pflags & TDP_DEADLKTREAT)))
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
#define LK_TRYOP(x) \
((x) & LK_NOWAIT)
#define LK_CAN_WITNESS(x) \
(((x) & LK_NOWITNESS) == 0 && !LK_TRYOP(x))
#define LK_TRYWIT(x) \
(LK_TRYOP(x) ? LOP_TRYLOCK : 0)
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#define LK_CAN_ADAPT(lk, f) \
(((lk)->lock_object.lo_flags & LK_ADAPTIVE) != 0 && \
((f) & LK_SLEEPFAIL) == 0)
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#define lockmgr_disowned(lk) \
(((lk)->lk_lock & ~(LK_FLAGMASK & ~LK_SHARE)) == LK_KERNPROC)
#define lockmgr_xlocked(lk) \
(((lk)->lk_lock & ~(LK_FLAGMASK & ~LK_SHARE)) == (uintptr_t)curthread)
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
static void assert_lockmgr(const struct lock_object *lock, int how);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#ifdef DDB
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
static void db_show_lockmgr(const struct lock_object *lock);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#endif
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
static void lock_lockmgr(struct lock_object *lock, int how);
Add the OpenSolaris dtrace lockstat provider. The lockstat provider adds probes for mutexes, reader/writer and shared/exclusive locks to gather contention statistics and other locking information for dtrace scripts, the lockstat(1M) command and other potential consumers. Reviewed by: attilio jhb jb Approved by: gnn (mentor)
2009-05-26 20:28:22 +00:00
#ifdef KDTRACE_HOOKS
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
static int owner_lockmgr(const struct lock_object *lock,
struct thread **owner);
Add the OpenSolaris dtrace lockstat provider. The lockstat provider adds probes for mutexes, reader/writer and shared/exclusive locks to gather contention statistics and other locking information for dtrace scripts, the lockstat(1M) command and other potential consumers. Reviewed by: attilio jhb jb Approved by: gnn (mentor)
2009-05-26 20:28:22 +00:00
#endif
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
static int unlock_lockmgr(struct lock_object *lock);
track lock class name in a way that doesn't break WITNESS
2006-11-13 05:41:46 +00:00
struct lock_class lock_class_lockmgr = {
Use C99-style struct member initialization for lock classes.
2007-03-09 16:19:34 +00:00
.lc_name = "lockmgr",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
.lc_flags = LC_RECURSABLE | LC_SLEEPABLE | LC_SLEEPLOCK | LC_UPGRADABLE,
Expand lock class with the "virtual" function lc_assert which will offer an unified way for all the lock primitives to express lock assertions. Currenty, lockmgrs and rmlocks don't have assertions, so just panic in that case. This will be a base for more callout improvements. Ok'ed by: jhb, jeff
2007-11-18 14:43:53 +00:00
.lc_assert = assert_lockmgr,
track lock class name in a way that doesn't break WITNESS
2006-11-13 05:41:46 +00:00
#ifdef DDB
Add two new function pointers 'lc_lock' and 'lc_unlock' to lock classes. These functions are intended to be used to drop a lock and then reacquire it when doing an sleep such as msleep(9). Both functions accept a 'struct lock_object *' as their first parameter. The 'lc_unlock' function returns an integer that is then passed as the second paramter to the subsequent 'lc_lock' function. This can be used to communicate state. For example, sx locks and rwlocks use this to indicate if the lock was share/read locked vs exclusive/write locked. Currently, spin mutexes and lockmgr locks do not provide working lc_lock and lc_unlock functions.
2007-03-09 16:27:11 +00:00
.lc_ddb_show = db_show_lockmgr,
track lock class name in a way that doesn't break WITNESS
2006-11-13 05:41:46 +00:00
#endif
Add two new function pointers 'lc_lock' and 'lc_unlock' to lock classes. These functions are intended to be used to drop a lock and then reacquire it when doing an sleep such as msleep(9). Both functions accept a 'struct lock_object *' as their first parameter. The 'lc_unlock' function returns an integer that is then passed as the second paramter to the subsequent 'lc_lock' function. This can be used to communicate state. For example, sx locks and rwlocks use this to indicate if the lock was share/read locked vs exclusive/write locked. Currently, spin mutexes and lockmgr locks do not provide working lc_lock and lc_unlock functions.
2007-03-09 16:27:11 +00:00
.lc_lock = lock_lockmgr,
Add the OpenSolaris dtrace lockstat provider. The lockstat provider adds probes for mutexes, reader/writer and shared/exclusive locks to gather contention statistics and other locking information for dtrace scripts, the lockstat(1M) command and other potential consumers. Reviewed by: attilio jhb jb Approved by: gnn (mentor)
2009-05-26 20:28:22 +00:00
.lc_unlock = unlock_lockmgr,
#ifdef KDTRACE_HOOKS
.lc_owner = owner_lockmgr,
#endif
track lock class name in a way that doesn't break WITNESS
2006-11-13 05:41:46 +00:00
};
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#ifdef ADAPTIVE_LOCKMGRS
static u_int alk_retries = 10;
static u_int alk_loops = 10000;
Mark all SYSCTL_NODEs static that have no corresponding SYSCTL_DECLs. The SYSCTL_NODE macro defines a list that stores all child-elements of that node. If there's no SYSCTL_DECL macro anywhere else, there's no reason why it shouldn't be static.
2011-11-07 15:43:11 +00:00
static SYSCTL_NODE(_debug, OID_AUTO, lockmgr, CTLFLAG_RD, NULL,
"lockmgr debugging");
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
SYSCTL_UINT(_debug_lockmgr, OID_AUTO, retries, CTLFLAG_RW, &alk_retries, 0, "");
SYSCTL_UINT(_debug_lockmgr, OID_AUTO, loops, CTLFLAG_RW, &alk_loops, 0, "");
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
static __inline struct thread *
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
lockmgr_xholder(const struct lock *lk)
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
{
uintptr_t x;
x = lk->lk_lock;
return ((x & LK_SHARE) ? NULL : (struct thread *)LK_HOLDER(x));
}
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
/*
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* It assumes sleepq_lock held and returns with this one unheld.
* It also assumes the generic interlock is sane and previously checked.
* If LK_INTERLOCK is specified the interlock is not reacquired after the
* sleep.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*/
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
static __inline int
sleeplk(struct lock *lk, u_int flags, struct lock_object *ilk,
const char *wmesg, int pri, int timo, int queue)
{
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
GIANT_DECLARE;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
struct lock_class *class;
int catch, error;
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
class = (flags & LK_INTERLOCK) ? LOCK_CLASS(ilk) : NULL;
The "if" semantic is not needed, just fix this.
2008-05-25 16:11:27 +00:00
catch = pri & PCATCH;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
pri &= PRIMASK;
error = 0;
LOCK_LOG3(lk, "%s: %p blocking on the %s sleepqueue", __func__, lk,
(queue == SQ_EXCLUSIVE_QUEUE) ? "exclusive" : "shared");
if (flags & LK_INTERLOCK)
class->lc_unlock(ilk);
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
if (queue == SQ_EXCLUSIVE_QUEUE && (flags & LK_SLEEPFAIL) != 0)
lk->lk_exslpfail++;
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
GIANT_SAVE();
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
sleepq_add(&lk->lock_object, NULL, wmesg, SLEEPQ_LK | (catch ?
SLEEPQ_INTERRUPTIBLE : 0), queue);
if ((flags & LK_TIMELOCK) && timo)
sleepq_set_timeout(&lk->lock_object, timo);
/*
* Decisional switch for real sleeping.
*/
if ((flags & LK_TIMELOCK) && timo && catch)
error = sleepq_timedwait_sig(&lk->lock_object, pri);
else if ((flags & LK_TIMELOCK) && timo)
error = sleepq_timedwait(&lk->lock_object, pri);
else if (catch)
error = sleepq_wait_sig(&lk->lock_object, pri);
else
sleepq_wait(&lk->lock_object, pri);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
GIANT_RESTORE();
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if ((flags & LK_SLEEPFAIL) && error == 0)
error = ENOLCK;
return (error);
}
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
static __inline int
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
wakeupshlk(struct lock *lk, const char *file, int line)
{
uintptr_t v, x;
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
u_int realexslp;
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
int queue, wakeup_swapper;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
TD_LOCKS_DEC(curthread);
TD_SLOCKS_DEC(curthread);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
WITNESS_UNLOCK(&lk->lock_object, 0, file, line);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, file, line);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
wakeup_swapper = 0;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
for (;;) {
x = lk->lk_lock;
/*
* If there is more than one shared lock held, just drop one
* and return.
*/
if (LK_SHARERS(x) > 1) {
When releasing a lockmgr held in shared way we need to use a write memory barrier in order to avoid, on architectures which doesn't have strong ordered writes, CPU instructions reordering. Diagnosed by: fabio
2009-10-03 15:02:55 +00:00
if (atomic_cmpset_rel_ptr(&lk->lk_lock, x,
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
x - LK_ONE_SHARER))
break;
continue;
}
/*
* If there are not waiters on the exclusive queue, drop the
* lock quickly.
*/
if ((x & LK_ALL_WAITERS) == 0) {
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
MPASS((x & ~LK_EXCLUSIVE_SPINNERS) ==
LK_SHARERS_LOCK(1));
When releasing a lockmgr held in shared way we need to use a write memory barrier in order to avoid, on architectures which doesn't have strong ordered writes, CPU instructions reordering. Diagnosed by: fabio
2009-10-03 15:02:55 +00:00
if (atomic_cmpset_rel_ptr(&lk->lk_lock, x, LK_UNLOCKED))
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
break;
continue;
}
/*
* We should have a sharer with waiters, so enter the hard
* path in order to handle wakeups correctly.
*/
sleepq_lock(&lk->lock_object);
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
x = lk->lk_lock & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
v = LK_UNLOCKED;
/*
* If the lock has exclusive waiters, give them preference in
* order to avoid deadlock with shared runners up.
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
* If interruptible sleeps left the exclusive queue empty
* avoid a starvation for the threads sleeping on the shared
* queue by giving them precedence and cleaning up the
* exclusive waiters bit anyway.
Tweak comments.
2010-01-07 01:19:01 +00:00
* Please note that lk_exslpfail count may be lying about
* the real number of waiters with the LK_SLEEPFAIL flag on
* because they may be used in conjuction with interruptible
Fix typos.
2010-01-07 01:24:09 +00:00
* sleeps so lk_exslpfail might be considered an 'upper limit'
* bound, including the edge cases.
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
*/
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
realexslp = sleepq_sleepcnt(&lk->lock_object,
SQ_EXCLUSIVE_QUEUE);
if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) {
if (lk->lk_exslpfail < realexslp) {
lk->lk_exslpfail = 0;
queue = SQ_EXCLUSIVE_QUEUE;
v |= (x & LK_SHARED_WAITERS);
} else {
lk->lk_exslpfail = 0;
LOCK_LOG2(lk,
"%s: %p has only LK_SLEEPFAIL sleepers",
__func__, lk);
LOCK_LOG2(lk,
"%s: %p waking up threads on the exclusive queue",
__func__, lk);
wakeup_swapper =
sleepq_broadcast(&lk->lock_object,
SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE);
queue = SQ_SHARED_QUEUE;
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
} else {
Exclusive waiters sleeping with LK_SLEEPFAIL on and using interruptible sleeps/timeout may have left spourious lk_exslpfail counts on, so clean it up even when accessing a shared queue acquisition, giving to lk_exslpfail the value of 'upper limit'. In the worst case scenario, infact (mixed interruptible sleep / LK_SLEEPFAIL waiters) what may happen is that both queues are awaken even if that's not necessary, but still no harm. Reported by: Lucius Windschuh <lwindschuh at googlemail dot com> Reviewed by: kib Tested by: pho, Lucius Windschuh <lwindschuh at googlemail dot com>
2010-01-07 00:47:50 +00:00
/*
* Exclusive waiters sleeping with LK_SLEEPFAIL on
* and using interruptible sleeps/timeout may have
* left spourious lk_exslpfail counts on, so clean
* it up anyway.
*/
lk->lk_exslpfail = 0;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
queue = SQ_SHARED_QUEUE;
}
When releasing a lockmgr held in shared way we need to use a write memory barrier in order to avoid, on architectures which doesn't have strong ordered writes, CPU instructions reordering. Diagnosed by: fabio
2009-10-03 15:02:55 +00:00
if (!atomic_cmpset_rel_ptr(&lk->lk_lock, LK_SHARERS_LOCK(1) | x,
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
v)) {
sleepq_release(&lk->lock_object);
continue;
}
LOCK_LOG3(lk, "%s: %p waking up threads on the %s queue",
__func__, lk, queue == SQ_SHARED_QUEUE ? "shared" :
"exclusive");
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
wakeup_swapper |= sleepq_broadcast(&lk->lock_object, SLEEPQ_LK,
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
0, queue);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
sleepq_release(&lk->lock_object);
break;
}
lock_profile_release_lock(&lk->lock_object);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
return (wakeup_swapper);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
}
static void
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
assert_lockmgr(const struct lock_object *lock, int what)
Expand lock class with the "virtual" function lc_assert which will offer an unified way for all the lock primitives to express lock assertions. Currenty, lockmgrs and rmlocks don't have assertions, so just panic in that case. This will be a base for more callout improvements. Ok'ed by: jhb, jeff
2007-11-18 14:43:53 +00:00
{
panic("lockmgr locks do not support assertions");
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
static void
Add two new function pointers 'lc_lock' and 'lc_unlock' to lock classes. These functions are intended to be used to drop a lock and then reacquire it when doing an sleep such as msleep(9). Both functions accept a 'struct lock_object *' as their first parameter. The 'lc_unlock' function returns an integer that is then passed as the second paramter to the subsequent 'lc_lock' function. This can be used to communicate state. For example, sx locks and rwlocks use this to indicate if the lock was share/read locked vs exclusive/write locked. Currently, spin mutexes and lockmgr locks do not provide working lc_lock and lc_unlock functions.
2007-03-09 16:27:11 +00:00
lock_lockmgr(struct lock_object *lock, int how)
{
panic("lockmgr locks do not support sleep interlocking");
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
static int
Add two new function pointers 'lc_lock' and 'lc_unlock' to lock classes. These functions are intended to be used to drop a lock and then reacquire it when doing an sleep such as msleep(9). Both functions accept a 'struct lock_object *' as their first parameter. The 'lc_unlock' function returns an integer that is then passed as the second paramter to the subsequent 'lc_lock' function. This can be used to communicate state. For example, sx locks and rwlocks use this to indicate if the lock was share/read locked vs exclusive/write locked. Currently, spin mutexes and lockmgr locks do not provide working lc_lock and lc_unlock functions.
2007-03-09 16:27:11 +00:00
unlock_lockmgr(struct lock_object *lock)
{
panic("lockmgr locks do not support sleep interlocking");
}
Add the OpenSolaris dtrace lockstat provider. The lockstat provider adds probes for mutexes, reader/writer and shared/exclusive locks to gather contention statistics and other locking information for dtrace scripts, the lockstat(1M) command and other potential consumers. Reviewed by: attilio jhb jb Approved by: gnn (mentor)
2009-05-26 20:28:22 +00:00
#ifdef KDTRACE_HOOKS
static int
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
owner_lockmgr(const struct lock_object *lock, struct thread **owner)
Add the OpenSolaris dtrace lockstat provider. The lockstat provider adds probes for mutexes, reader/writer and shared/exclusive locks to gather contention statistics and other locking information for dtrace scripts, the lockstat(1M) command and other potential consumers. Reviewed by: attilio jhb jb Approved by: gnn (mentor)
2009-05-26 20:28:22 +00:00
{
panic("lockmgr locks do not support owner inquiring");
}
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
void
lockinit(struct lock *lk, int pri, const char *wmesg, int timo, int flags)
{
int iflags;
Change the M_NAMEI allocations to use the zone allocator. This change plus the previous changes to use the zone allocator decrease the useage of malloc by half. The Zone allocator will be upgradeable to be able to use per CPU-pools, and has more intelligent usage of SPLs. Additionally, it has reasonable stats gathering capabilities, while making most calls inline.
1997-09-21 04:24:27 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
MPASS((flags & ~LK_INIT_MASK) == 0);
* Change the scope of the ASSERT_ATOMIC_LOAD() from a generic check to a pointer-fetching specific operation check. Consequently, rename the operation ASSERT_ATOMIC_LOAD_PTR(). * Fix the implementation of ASSERT_ATOMIC_LOAD_PTR() by checking directly alignment on the word boundry, for all the given specific architectures. That's a bit too strict for some common case, but it assures safety. * Add a comment explaining the scope of the macro * Add a new stub in the lockmgr specific implementation Tested by: marcel (initial version), marius Reviewed by: rwatson, jhb (comment specific review) Approved by: re (kib)
2009-08-17 16:17:21 +00:00
ASSERT_ATOMIC_LOAD_PTR(lk->lk_lock,
("%s: lockmgr not aligned for %s: %p", __func__, wmesg,
&lk->lk_lock));
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Handle lock recursion differenty by always checking against LO_RECURSABLE instead the lock own flag itself. Tested by: pho
2009-06-02 13:03:35 +00:00
iflags = LO_SLEEPABLE | LO_UPGRADABLE;
if (flags & LK_CANRECURSE)
iflags |= LO_RECURSABLE;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if ((flags & LK_NODUP) == 0)
iflags |= LO_DUPOK;
if (flags & LK_NOPROFILE)
iflags |= LO_NOPROFILE;
if ((flags & LK_NOWITNESS) == 0)
iflags |= LO_WITNESS;
if (flags & LK_QUIET)
iflags |= LO_QUIET;
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
iflags |= flags & (LK_ADAPTIVE | LK_NOSHARE);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lk_lock = LK_UNLOCKED;
lk->lk_recurse = 0;
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
lk->lk_exslpfail = 0;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lk_timo = timo;
lk->lk_pri = pri;
lock_init(&lk->lock_object, &lock_class_lockmgr, wmesg, NULL, iflags);
STACK_ZERO(lk);
Fix kern_lock so that it will work. Additionally, clean-up some of the VM systems usage of the kernel lock (lockmgr) code. This is a first pass implementation, and is expected to evolve as needed. The API for the lock manager code has not changed, but the underlying implementation has changed significantly. This change should not materially affect our current SMP or UP code without non-standard parameters being used.
1997-08-18 02:06:35 +00:00
}
Add dedicated routines to toggle lockmgr flags such as LK_NOSHARE and LK_CANRECURSE after a lock is created. Use them to implement macros that otherwise manipulated the flags directly. Assert that the associated lockmgr lock is exclusively locked by the current thread when manipulating these flags to ensure the flag updates are safe. This last change required some minor shuffling in a few filesystems to exclusively lock a brand new vnode slightly earlier. Reviewed by: kib MFC after: 3 days
2010-08-20 19:46:50 +00:00
/*
* XXX: Gross hacks to manipulate external lock flags after
* initialization. Used for certain vnode and buf locks.
*/
void
lockallowshare(struct lock *lk)
{
lockmgr_assert(lk, KA_XLOCKED);
lk->lock_object.lo_flags &= ~LK_NOSHARE;
}
void
lockallowrecurse(struct lock *lk)
{
lockmgr_assert(lk, KA_XLOCKED);
lk->lock_object.lo_flags |= LO_RECURSABLE;
}
void
lockdisablerecurse(struct lock *lk)
{
lockmgr_assert(lk, KA_XLOCKED);
lk->lock_object.lo_flags &= ~LO_RECURSABLE;
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
void
lockdestroy(struct lock *lk)
{
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
KASSERT(lk->lk_lock == LK_UNLOCKED, ("lockmgr still held"));
KASSERT(lk->lk_recurse == 0, ("lockmgr still recursed"));
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
KASSERT(lk->lk_exslpfail == 0, ("lockmgr still exclusive waiters"));
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lock_destroy(&lk->lock_object);
Fix kern_lock so that it will work. Additionally, clean-up some of the VM systems usage of the kernel lock (lockmgr) code. This is a first pass implementation, and is expected to evolve as needed. The API for the lock manager code has not changed, but the underlying implementation has changed significantly. This change should not materially affect our current SMP or UP code without non-standard parameters being used.
1997-08-18 02:06:35 +00:00
}
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
int
__lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk,
const char *wmesg, int pri, int timo, const char *file, int line)
- Do not use APAUSE if LK_INTERLOCK is set. We lose synchronization if the lockmgr interlock is dropped after the caller's interlock is dropped. - Change some lockmgr KTRs to be slightly more helpful. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:20:59 +00:00
{
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
GIANT_DECLARE;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
struct lock_class *class;
- Introduce lockmgr_args() in the lockmgr space. This function performs the same operation of lockmgr() but accepting a custom wmesg, prio and timo for the particular lock instance, overriding default values lkp->lk_wmesg, lkp->lk_prio and lkp->lk_timo. - Use lockmgr_args() in order to implement BUF_TIMELOCK() - Cleanup BUF_LOCK() - Remove LK_INTERNAL as it is nomore used in the lockmgr namespace Tested by: Andrea Barberio <insomniac at slackware dot it>
2008-02-15 21:04:36 +00:00
const char *iwmesg;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
uintptr_t tid, v, x;
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
u_int op, realexslp;
- Wrap lock profiling state variables in #ifdef LOCK_PROFILING blocks.
2009-03-15 08:03:54 +00:00
int error, ipri, itimo, queue, wakeup_swapper;
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#ifdef ADAPTIVE_LOCKMGRS
volatile struct thread *owner;
u_int i, spintries = 0;
#endif
- Introduce lockmgr_args() in the lockmgr space. This function performs the same operation of lockmgr() but accepting a custom wmesg, prio and timo for the particular lock instance, overriding default values lkp->lk_wmesg, lkp->lk_prio and lkp->lk_timo. - Use lockmgr_args() in order to implement BUF_TIMELOCK() - Cleanup BUF_LOCK() - Remove LK_INTERNAL as it is nomore used in the lockmgr namespace Tested by: Andrea Barberio <insomniac at slackware dot it>
2008-02-15 21:04:36 +00:00
- Slightly restructure acquire() so I can add more ktr information and an assert to help find two strange bugs. - Remove some nearby spls.
2005-04-03 11:49:02 +00:00
error = 0;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
tid = (uintptr_t)curthread;
op = (flags & LK_TYPE_MASK);
iwmesg = (wmesg == LK_WMESG_DEFAULT) ? lk->lock_object.lo_name : wmesg;
ipri = (pri == LK_PRIO_DEFAULT) ? lk->lk_pri : pri;
itimo = (timo == LK_TIMO_DEFAULT) ? lk->lk_timo : timo;
MPASS((flags & ~LK_TOTAL_MASK) == 0);
- Remove a stale comment. - Add an extra assertion in order to catch malformed requested operations.
2008-04-12 13:56:17 +00:00
KASSERT((op & (op - 1)) == 0,
("%s: Invalid requested operation @ %s:%d", __func__, file, line));
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
KASSERT((flags & (LK_NOWAIT | LK_SLEEPFAIL)) == 0 ||
(op != LK_DOWNGRADE && op != LK_RELEASE),
("%s: Invalid flags in regard of the operation desired @ %s:%d",
__func__, file, line));
KASSERT((flags & LK_INTERLOCK) == 0 || ilk != NULL,
("%s: LK_INTERLOCK passed without valid interlock @ %s:%d",
__func__, file, line));
Improve check coverage about idle threads. Idle threads are not allowed to acquire any lock but spinlocks. Deny any attempt to do so by panicing at the locking operation when INVARIANTS is on. Then, remove the check on blocking on a turnstile. The check in sleepqueues is left because they are not allowed to use tsleep() either which could happen still. Reviewed by: bde, jhb, kib MFC after: 1 week
2012-09-12 22:10:53 +00:00
KASSERT(!TD_IS_IDLETHREAD(curthread),
("%s: idle thread %p on lockmgr %s @ %s:%d", __func__, curthread,
lk->lock_object.lo_name, file, line));
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
class = (flags & LK_INTERLOCK) ? LOCK_CLASS(ilk) : NULL;
if (panicstr != NULL) {
if (flags & LK_INTERLOCK)
class->lc_unlock(ilk);
return (0);
Fix kern_lock so that it will work. Additionally, clean-up some of the VM systems usage of the kernel lock (lockmgr) code. This is a first pass implementation, and is expected to evolve as needed. The API for the lock manager code has not changed, but the underlying implementation has changed significantly. This change should not materially affect our current SMP or UP code without non-standard parameters being used.
1997-08-18 02:06:35 +00:00
}
Fix the LK_NOSHARE lockmgr flag interaction with LK_UPGRADE and LK_DOWNGRADE lock ops. Namely, the ops should be NOP since LK_NOSHARE locks are always exclusive. Reported by: rmacklem Reviewed by: attilio Tested by: pho Approved by: re (kensmith) MFC after: 1 week
2011-08-01 19:07:03 +00:00
if (lk->lock_object.lo_flags & LK_NOSHARE) {
switch (op) {
case LK_SHARED:
op = LK_EXCLUSIVE;
break;
case LK_UPGRADE:
case LK_DOWNGRADE:
_lockmgr_assert(lk, KA_XLOCKED | KA_NOTRECURSED,
file, line);
return (0);
}
}
MUTEX_PROFILING has been generalized to LOCK_PROFILING. We now profile wait (time waited to acquire) and hold times for *all* kernel locks. If the architecture has a system synchronized TSC, the profiling code will use that - thereby minimizing profiling overhead. Large chunks of profiling code have been moved out of line, the overhead measured on the T1 for when it is compiled in but not enabled is < 1%. Approved by: scottl (standing in for mentor rwatson) Reviewed by: des and jhb
2006-11-11 03:18:07 +00:00
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
wakeup_swapper = 0;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
switch (op) {
case LK_SHARED:
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
if (LK_CAN_WITNESS(flags))
WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER,
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
file, line, ilk);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
for (;;) {
x = lk->lk_lock;
In order to avoid a huge class of deadlocks (in particular in interactions with the interlock), owner of the lock should be only curthread or at least, for its limited usage, NULL which identifies LK_KERNPROC. The thread "extra argument" for the lockmgr interface is going to be removed in the near future, but for the moment, just let kernel run for some days with this check on in order to find potential deadlocking places around the kernel and fix them.
2007-12-27 22:56:57 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* If no other thread has an exclusive lock, or
* no exclusive waiter is present, bump the count of
* sharers. Since we have to preserve the state of
* waiters, if we fail to acquire the shared lock
* loop back and retry.
*/
if (LK_CAN_SHARE(x)) {
if (atomic_cmpset_acq_ptr(&lk->lk_lock, x,
x + LK_ONE_SHARER))
break;
continue;
}
Add software PMC support. New kernel events can be added at various location for sampling or counting. This will for example allow easy system profiling whatever the processor is with known tools like pmcstat(8). Simultaneous usage of software PMC and hardware PMC is possible, for example looking at the lock acquire failure, page fault while sampling on instructions. Sponsored by: NETASQ MFC after: 1 month
2012-03-28 20:58:30 +00:00
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lock_profile_obtain_lock_failed(&lk->lock_object,
&contested, &waittime);
Fix kern_lock so that it will work. Additionally, clean-up some of the VM systems usage of the kernel lock (lockmgr) code. This is a first pass implementation, and is expected to evolve as needed. The API for the lock manager code has not changed, but the underlying implementation has changed significantly. This change should not materially affect our current SMP or UP code without non-standard parameters being used.
1997-08-18 02:06:35 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
s/alredy/already/ in the comments and the log message.
2008-07-25 11:22:25 +00:00
* If the lock is already held by curthread in
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* exclusive way avoid a deadlock.
*/
if (LK_HOLDER(x) == tid) {
LOCK_LOG2(lk,
s/alredy/already/ in the comments and the log message.
2008-07-25 11:22:25 +00:00
"%s: %p already held in exclusive mode",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
__func__, lk);
error = EDEADLK;
break;
}
- Do not use APAUSE if LK_INTERLOCK is set. We lose synchronization if the lockmgr interlock is dropped after the caller's interlock is dropped. - Change some lockmgr KTRs to be slightly more helpful. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:20:59 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* If the lock is expected to not sleep just give up
* and return.
*/
if (LK_TRYOP(flags)) {
LOCK_LOG2(lk, "%s: %p fails the try operation",
__func__, lk);
error = EBUSY;
break;
}
Fix kern_lock so that it will work. Additionally, clean-up some of the VM systems usage of the kernel lock (lockmgr) code. This is a first pass implementation, and is expected to evolve as needed. The API for the lock manager code has not changed, but the underlying implementation has changed significantly. This change should not materially affect our current SMP or UP code without non-standard parameters being used.
1997-08-18 02:06:35 +00:00
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#ifdef ADAPTIVE_LOCKMGRS
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
Fix some bugs related to adaptive spinning: In the lockmgr support: - GIANT_RESTORE() is just called when the sleep finishes, so the current code can ends up into a giant unlock problem. Fix it by appropriately call GIANT_RESTORE() when needed. Note that this is not exactly ideal because for any interation of the adaptive spinning we drop and restore Giant, but the overhead should be not a factor. - In the lock held in exclusive mode case, after the adaptive spinning is brought to completition, we should just retry to acquire the lock instead to fallthrough. Fix that. - Fix a style nit In the sx support: - Call GIANT_SAVE() before than looping. This saves some overhead because in the current code GIANT_SAVE() is called several times. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-02 17:33:51 +00:00
* changes. We need a double-state handle here
* because for a failed acquisition the lock can be
* either held in exclusive mode or shared mode
* (for the writer starvation avoidance technique).
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
*/
if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 &&
LK_HOLDER(x) != LK_KERNPROC) {
owner = (struct thread *)LK_HOLDER(x);
if (LOCK_LOG_TEST(&lk->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, lk, owner);
/*
* If we are holding also an interlock drop it
* in order to avoid a deadlock if the lockmgr
* owner is adaptively spinning on the
* interlock itself.
*/
if (flags & LK_INTERLOCK) {
class->lc_unlock(ilk);
flags &= ~LK_INTERLOCK;
}
GIANT_SAVE();
while (LK_HOLDER(lk->lk_lock) ==
(uintptr_t)owner && TD_IS_RUNNING(owner))
cpu_spinwait();
Fix some bugs related to adaptive spinning: In the lockmgr support: - GIANT_RESTORE() is just called when the sleep finishes, so the current code can ends up into a giant unlock problem. Fix it by appropriately call GIANT_RESTORE() when needed. Note that this is not exactly ideal because for any interation of the adaptive spinning we drop and restore Giant, but the overhead should be not a factor. - In the lock held in exclusive mode case, after the adaptive spinning is brought to completition, we should just retry to acquire the lock instead to fallthrough. Fix that. - Fix a style nit In the sx support: - Call GIANT_SAVE() before than looping. This saves some overhead because in the current code GIANT_SAVE() is called several times. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-02 17:33:51 +00:00
GIANT_RESTORE();
continue;
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
} else if (LK_CAN_ADAPT(lk, flags) &&
Fix some bugs related to adaptive spinning: In the lockmgr support: - GIANT_RESTORE() is just called when the sleep finishes, so the current code can ends up into a giant unlock problem. Fix it by appropriately call GIANT_RESTORE() when needed. Note that this is not exactly ideal because for any interation of the adaptive spinning we drop and restore Giant, but the overhead should be not a factor. - In the lock held in exclusive mode case, after the adaptive spinning is brought to completition, we should just retry to acquire the lock instead to fallthrough. Fix that. - Fix a style nit In the sx support: - Call GIANT_SAVE() before than looping. This saves some overhead because in the current code GIANT_SAVE() is called several times. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-02 17:33:51 +00:00
(x & LK_SHARE) != 0 && LK_SHARERS(x) &&
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
spintries < alk_retries) {
if (flags & LK_INTERLOCK) {
class->lc_unlock(ilk);
flags &= ~LK_INTERLOCK;
}
GIANT_SAVE();
spintries++;
for (i = 0; i < alk_loops; i++) {
if (LOCK_LOG_TEST(&lk->lock_object, 0))
CTR4(KTR_LOCK,
"%s: shared spinning on %p with %u and %u",
__func__, lk, spintries, i);
x = lk->lk_lock;
if ((x & LK_SHARE) == 0 ||
LK_CAN_SHARE(x) != 0)
break;
cpu_spinwait();
}
Fix some bugs related to adaptive spinning: In the lockmgr support: - GIANT_RESTORE() is just called when the sleep finishes, so the current code can ends up into a giant unlock problem. Fix it by appropriately call GIANT_RESTORE() when needed. Note that this is not exactly ideal because for any interation of the adaptive spinning we drop and restore Giant, but the overhead should be not a factor. - In the lock held in exclusive mode case, after the adaptive spinning is brought to completition, we should just retry to acquire the lock instead to fallthrough. Fix that. - Fix a style nit In the sx support: - Call GIANT_SAVE() before than looping. This saves some overhead because in the current code GIANT_SAVE() is called several times. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-02 17:33:51 +00:00
GIANT_RESTORE();
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
if (i != alk_loops)
continue;
}
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* Acquire the sleepqueue chain lock because we
* probabilly will need to manipulate waiters flags.
*/
sleepq_lock(&lk->lock_object);
x = lk->lk_lock;
- Add an interlock argument to BUF_LOCK and BUF_TIMELOCK. - Remove the buftimelock mutex and acquire the buf's interlock to protect these fields instead. - Hold the vnode interlock while locking bufs on the clean/dirty queues. This reduces some cases from one BUF_LOCK with a LK_NOWAIT and another BUF_LOCK with a LK_TIMEFAIL to a single lock. Reviewed by: arch, mckusick
2003-02-25 03:37:48 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* if the lock can be acquired in shared mode, try
* again.
*/
if (LK_CAN_SHARE(x)) {
sleepq_release(&lk->lock_object);
continue;
}
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#ifdef ADAPTIVE_LOCKMGRS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owner) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 &&
LK_HOLDER(x) != LK_KERNPROC) {
owner = (struct thread *)LK_HOLDER(x);
if (TD_IS_RUNNING(owner)) {
sleepq_release(&lk->lock_object);
continue;
}
}
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* Try to set the LK_SHARED_WAITERS flag. If we fail,
* loop back and retry.
*/
if ((x & LK_SHARED_WAITERS) == 0) {
if (!atomic_cmpset_acq_ptr(&lk->lk_lock, x,
x | LK_SHARED_WAITERS)) {
sleepq_release(&lk->lock_object);
continue;
}
LOCK_LOG2(lk, "%s: %p set shared waiters flag",
__func__, lk);
}
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* As far as we have been unable to acquire the
* shared lock and the shared waiters flag is set,
* we will sleep.
*/
error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo,
SQ_SHARED_QUEUE);
flags &= ~LK_INTERLOCK;
if (error) {
LOCK_LOG3(lk,
"%s: interrupted sleep for %p with %d",
__func__, lk, error);
break;
}
LOCK_LOG2(lk, "%s: %p resuming from the sleep queue",
__func__, lk);
}
if (error == 0) {
lock_profile_obtain_lock_success(&lk->lock_object,
contested, waittime, file, line);
LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, file,
Add WITNESS support to lockmgr locking primitive. This support tries to be as parallel as possible with other locking primitives, but there are differences; more specifically: - The base witness support is alredy equipped for allowing lock duplication acquisition as lockmgr rely on this. - In the case of lockmgr_disown() the lock result unlocked by witness even if it is still held by the "kernel context" - In the case of upgrading we can have 3 different situations: * Total unlocking of the shared lock and nothing else * Real witness upgrade if the owner is the first upgrader * Shared unlocking and exclusive locking if the owner is not the first upgrade but it is still allowed to upgrade - LK_DRAIN is basically handled like an exclusive acquisition Additively new options LK_NODUP and LK_NOWITNESS can now be used with lockinit(): LK_NOWITNESS disables WITNESS for the specified lock while LK_NODUP enable duplicated locks tracking. This will require manpages update and a __FreeBSD_version bumping (addressed by further commits). This patch also fixes a problem occurring if a lockmgr is held in exclusive mode and the same owner try to acquire it in shared mode: currently there is a spourious shared locking acquisition while what we really want is a lock downgrade. Probabilly, this situation can be better served with a EDEADLK failing errno return. Side note: first testing on this patch alredy reveleated several LORs reported, so please expect LORs cascades until resolved. NTFS also is reported broken by WITNESS introduction. BTW, NTFS is exposing a lock leak which needs to be fixed, and this patch can help it out if rightly tweaked. Tested by: kris, yar, Scot Hetzel <swhetzel at gmail dot com>
2008-02-06 00:37:14 +00:00
line);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
WITNESS_LOCK(&lk->lock_object, LK_TRYWIT(flags), file,
line);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
TD_LOCKS_INC(curthread);
TD_SLOCKS_INC(curthread);
STACK_SAVE(lk);
}
break;
case LK_UPGRADE:
_lockmgr_assert(lk, KA_SLOCKED, file, line);
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
v = lk->lk_lock;
x = v & LK_ALL_WAITERS;
v &= LK_EXCLUSIVE_SPINNERS;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
This solves a deadlock that can occur when read()ing into a file-mmap() space. When doing this, it is possible to for another process to attempt to get an exclusive lock on the vnode and deadlock the mmap/read combination when the uiomove() call tries to obtain a second shared lock on the vnode. There is still a potential deadlock situation with write()/mmap(). Submitted by: Matt Dillon <dillon@freebsd.org> Reviewed by: Luoqi Chen <luoqi@freebsd.org> Delimmitted by tag PRE_MATT_MMAP_LOCK and POST_MATT_MMAP_LOCK in kern/kern_lock.c kern/kern_subr.c
1999-03-12 03:09:29 +00:00
/*
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* Try to switch from one shared lock to an exclusive one.
* We need to preserve waiters flags during the operation.
This solves a deadlock that can occur when read()ing into a file-mmap() space. When doing this, it is possible to for another process to attempt to get an exclusive lock on the vnode and deadlock the mmap/read combination when the uiomove() call tries to obtain a second shared lock on the vnode. There is still a potential deadlock situation with write()/mmap(). Submitted by: Matt Dillon <dillon@freebsd.org> Reviewed by: Luoqi Chen <luoqi@freebsd.org> Delimmitted by tag PRE_MATT_MMAP_LOCK and POST_MATT_MMAP_LOCK in kern/kern_lock.c kern/kern_subr.c
1999-03-12 03:09:29 +00:00
*/
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
if (atomic_cmpset_ptr(&lk->lk_lock, LK_SHARERS_LOCK(1) | x | v,
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
tid | x)) {
LOCK_LOG_LOCK("XUPGRADE", &lk->lock_object, 0, 0, file,
line);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
WITNESS_UPGRADE(&lk->lock_object, LOP_EXCLUSIVE |
LK_TRYWIT(flags), file, line);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
TD_SLOCKS_DEC(curthread);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
break;
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
/*
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* We have been unable to succeed in upgrading, so just
* give up the shared lock.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*/
Use |= rather than += when aggregrating requests to wakeup the swapper. What we really want is an inclusive or of all the requests, and += can in theory roll over to 0.
2008-08-22 16:14:23 +00:00
wakeup_swapper |= wakeupshlk(lk, file, line);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/* FALLTHROUGH */
case LK_EXCLUSIVE:
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
if (LK_CAN_WITNESS(flags))
WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER |
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
LOP_EXCLUSIVE, file, line, ilk);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
/*
s/alredy/already/ in the comments and the log message.
2008-07-25 11:22:25 +00:00
* If curthread already holds the lock and this one is
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* allowed to recurse, simply recurse on it.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*/
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (lockmgr_xlocked(lk)) {
if ((flags & LK_CANRECURSE) == 0 &&
Handle lock recursion differenty by always checking against LO_RECURSABLE instead the lock own flag itself. Tested by: pho
2009-06-02 13:03:35 +00:00
(lk->lock_object.lo_flags & LO_RECURSABLE) == 0) {
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* If the lock is expected to not panic just
* give up and return.
*/
if (LK_TRYOP(flags)) {
LOCK_LOG2(lk,
"%s: %p fails the try operation",
__func__, lk);
error = EBUSY;
break;
}
if (flags & LK_INTERLOCK)
class->lc_unlock(ilk);
panic("%s: recursing on non recursive lockmgr %s @ %s:%d\n",
__func__, iwmesg, file, line);
}
lk->lk_recurse++;
LOCK_LOG2(lk, "%s: %p recursing", __func__, lk);
LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0,
lk->lk_recurse, file, line);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE |
LK_TRYWIT(flags), file, line);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
TD_LOCKS_INC(curthread);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
break;
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
while (!atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED,
tid)) {
Add software PMC support. New kernel events can be added at various location for sampling or counting. This will for example allow easy system profiling whatever the processor is with known tools like pmcstat(8). Simultaneous usage of software PMC and hardware PMC is possible, for example looking at the lock acquire failure, page fault while sampling on instructions. Sponsored by: NETASQ MFC after: 1 month
2012-03-28 20:58:30 +00:00
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lock_profile_obtain_lock_failed(&lk->lock_object,
&contested, &waittime);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
/*
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* If the lock is expected to not sleep just give up
* and return.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*/
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (LK_TRYOP(flags)) {
LOCK_LOG2(lk, "%s: %p fails the try operation",
__func__, lk);
error = EBUSY;
break;
}
Some kern_lock code improvements. Add missing wakeup, and enable disabling some diagnostics when memory or speed is at a premium.
1998-03-07 19:25:34 +00:00
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#ifdef ADAPTIVE_LOCKMGRS
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
* changes.
*/
x = lk->lk_lock;
if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 &&
LK_HOLDER(x) != LK_KERNPROC) {
owner = (struct thread *)LK_HOLDER(x);
if (LOCK_LOG_TEST(&lk->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, lk, owner);
/*
* If we are holding also an interlock drop it
* in order to avoid a deadlock if the lockmgr
* owner is adaptively spinning on the
* interlock itself.
*/
if (flags & LK_INTERLOCK) {
class->lc_unlock(ilk);
flags &= ~LK_INTERLOCK;
}
GIANT_SAVE();
while (LK_HOLDER(lk->lk_lock) ==
(uintptr_t)owner && TD_IS_RUNNING(owner))
cpu_spinwait();
Fix some bugs related to adaptive spinning: In the lockmgr support: - GIANT_RESTORE() is just called when the sleep finishes, so the current code can ends up into a giant unlock problem. Fix it by appropriately call GIANT_RESTORE() when needed. Note that this is not exactly ideal because for any interation of the adaptive spinning we drop and restore Giant, but the overhead should be not a factor. - In the lock held in exclusive mode case, after the adaptive spinning is brought to completition, we should just retry to acquire the lock instead to fallthrough. Fix that. - Fix a style nit In the sx support: - Call GIANT_SAVE() before than looping. This saves some overhead because in the current code GIANT_SAVE() is called several times. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-02 17:33:51 +00:00
GIANT_RESTORE();
continue;
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
} else if (LK_CAN_ADAPT(lk, flags) &&
(x & LK_SHARE) != 0 && LK_SHARERS(x) &&
spintries < alk_retries) {
if ((x & LK_EXCLUSIVE_SPINNERS) == 0 &&
!atomic_cmpset_ptr(&lk->lk_lock, x,
x | LK_EXCLUSIVE_SPINNERS))
continue;
if (flags & LK_INTERLOCK) {
class->lc_unlock(ilk);
flags &= ~LK_INTERLOCK;
}
GIANT_SAVE();
spintries++;
for (i = 0; i < alk_loops; i++) {
if (LOCK_LOG_TEST(&lk->lock_object, 0))
CTR4(KTR_LOCK,
"%s: shared spinning on %p with %u and %u",
__func__, lk, spintries, i);
if ((lk->lk_lock &
LK_EXCLUSIVE_SPINNERS) == 0)
break;
cpu_spinwait();
}
Fix some bugs related to adaptive spinning: In the lockmgr support: - GIANT_RESTORE() is just called when the sleep finishes, so the current code can ends up into a giant unlock problem. Fix it by appropriately call GIANT_RESTORE() when needed. Note that this is not exactly ideal because for any interation of the adaptive spinning we drop and restore Giant, but the overhead should be not a factor. - In the lock held in exclusive mode case, after the adaptive spinning is brought to completition, we should just retry to acquire the lock instead to fallthrough. Fix that. - Fix a style nit In the sx support: - Call GIANT_SAVE() before than looping. This saves some overhead because in the current code GIANT_SAVE() is called several times. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2009-09-02 17:33:51 +00:00
GIANT_RESTORE();
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
if (i != alk_loops)
continue;
}
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* Acquire the sleepqueue chain lock because we
* probabilly will need to manipulate waiters flags.
*/
sleepq_lock(&lk->lock_object);
x = lk->lk_lock;
/*
* if the lock has been released while we spun on
* the sleepqueue chain lock just try again.
*/
if (x == LK_UNLOCKED) {
sleepq_release(&lk->lock_object);
continue;
}
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
#ifdef ADAPTIVE_LOCKMGRS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owner) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 &&
LK_HOLDER(x) != LK_KERNPROC) {
owner = (struct thread *)LK_HOLDER(x);
if (TD_IS_RUNNING(owner)) {
sleepq_release(&lk->lock_object);
continue;
}
}
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* The lock can be in the state where there is a
* pending queue of waiters, but still no owner.
* This happens when the lock is contested and an
* owner is going to claim the lock.
* If curthread is the one successfully acquiring it
* claim lock ownership and return, preserving waiters
* flags.
*/
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS);
if ((x & ~v) == LK_UNLOCKED) {
v &= ~LK_EXCLUSIVE_SPINNERS;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (atomic_cmpset_acq_ptr(&lk->lk_lock, x,
tid | v)) {
sleepq_release(&lk->lock_object);
LOCK_LOG2(lk,
"%s: %p claimed by a new writer",
__func__, lk);
break;
}
sleepq_release(&lk->lock_object);
continue;
Reintroduce slightly modified patch from kern/69964. Check for LK_HAVE_EXL in both acquire invocations. MFC after: 5 days
2004-08-27 01:41:28 +00:00
}
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
/*
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* Try to set the LK_EXCLUSIVE_WAITERS flag. If we
* fail, loop back and retry.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*/
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if ((x & LK_EXCLUSIVE_WAITERS) == 0) {
if (!atomic_cmpset_ptr(&lk->lk_lock, x,
x | LK_EXCLUSIVE_WAITERS)) {
sleepq_release(&lk->lock_object);
continue;
}
LOCK_LOG2(lk, "%s: %p set excl waiters flag",
__func__, lk);
}
/*
* As far as we have been unable to acquire the
* exclusive lock and the exclusive waiters flag
* is set, we will sleep.
*/
error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo,
SQ_EXCLUSIVE_QUEUE);
flags &= ~LK_INTERLOCK;
if (error) {
LOCK_LOG3(lk,
"%s: interrupted sleep for %p with %d",
__func__, lk, error);
When requesting an exclusive lock with LK_NOWAIT, do not panic if LK_RECURSIVE is not set, as we will simply return that the lock is busy and not actually deadlock. This allows processes to use polling locks against buffers that they may already hold exclusively locked.
1999-06-28 07:54:58 +00:00
break;
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
LOCK_LOG2(lk, "%s: %p resuming from the sleep queue",
__func__, lk);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (error == 0) {
lock_profile_obtain_lock_success(&lk->lock_object,
contested, waittime, file, line);
LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0,
lk->lk_recurse, file, line);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE |
LK_TRYWIT(flags), file, line);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
TD_LOCKS_INC(curthread);
STACK_SAVE(lk);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
break;
case LK_DOWNGRADE:
_lockmgr_assert(lk, KA_XLOCKED | KA_NOTRECURSED, file, line);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
LOCK_LOG_LOCK("XDOWNGRADE", &lk->lock_object, 0, 0, file, line);
WITNESS_DOWNGRADE(&lk->lock_object, 0, file, line);
TD_SLOCKS_INC(curthread);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
/*
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* In order to preserve waiters flags, just spin.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*/
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
for (;;) {
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
x = lk->lk_lock;
MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0);
x &= LK_ALL_WAITERS;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x,
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
LK_SHARERS_LOCK(1) | x))
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
break;
cpu_spinwait();
}
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
break;
case LK_RELEASE:
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
_lockmgr_assert(lk, KA_LOCKED, file, line);
x = lk->lk_lock;
if ((x & LK_SHARE) == 0) {
/*
* As first option, treact the lock as if it has not
* any waiter.
* Fix-up the tid var if the lock has been disowned.
*/
if (LK_HOLDER(x) == LK_KERNPROC)
tid = LK_KERNPROC;
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
else {
WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE,
file, line);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
TD_LOCKS_DEC(curthread);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
LOCK_LOG_LOCK("XUNLOCK", &lk->lock_object, 0,
lk->lk_recurse, file, line);
/*
* The lock is held in exclusive mode.
* If the lock is recursed also, then unrecurse it.
*/
if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) {
LOCK_LOG2(lk, "%s: %p unrecursing", __func__,
lk);
lk->lk_recurse--;
break;
Add WITNESS support to lockmgr locking primitive. This support tries to be as parallel as possible with other locking primitives, but there are differences; more specifically: - The base witness support is alredy equipped for allowing lock duplication acquisition as lockmgr rely on this. - In the case of lockmgr_disown() the lock result unlocked by witness even if it is still held by the "kernel context" - In the case of upgrading we can have 3 different situations: * Total unlocking of the shared lock and nothing else * Real witness upgrade if the owner is the first upgrader * Shared unlocking and exclusive locking if the owner is not the first upgrade but it is still allowed to upgrade - LK_DRAIN is basically handled like an exclusive acquisition Additively new options LK_NODUP and LK_NOWITNESS can now be used with lockinit(): LK_NOWITNESS disables WITNESS for the specified lock while LK_NODUP enable duplicated locks tracking. This will require manpages update and a __FreeBSD_version bumping (addressed by further commits). This patch also fixes a problem occurring if a lockmgr is held in exclusive mode and the same owner try to acquire it in shared mode: currently there is a spourious shared locking acquisition while what we really want is a lock downgrade. Probabilly, this situation can be better served with a EDEADLK failing errno return. Side note: first testing on this patch alredy reveleated several LORs reported, so please expect LORs cascades until resolved. NTFS also is reported broken by WITNESS introduction. BTW, NTFS is exposing a lock leak which needs to be fixed, and this patch can help it out if rightly tweaked. Tested by: kris, yar, Scot Hetzel <swhetzel at gmail dot com>
2008-02-06 00:37:14 +00:00
}
- Call lock_profile_release when we're transitioning a lock to be owned by LK_KERNPROC. Discussed with: attilio
2009-03-14 11:43:02 +00:00
if (tid != LK_KERNPROC)
lock_profile_release_lock(&lk->lock_object);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid,
LK_UNLOCKED))
break;
sleepq_lock(&lk->lock_object);
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
x = lk->lk_lock;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
v = LK_UNLOCKED;
/*
* If the lock has exclusive waiters, give them
* preference in order to avoid deadlock with
* shared runners up.
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
* If interruptible sleeps left the exclusive queue
* empty avoid a starvation for the threads sleeping
* on the shared queue by giving them precedence
* and cleaning up the exclusive waiters bit anyway.
Tweak comments.
2010-01-07 01:19:01 +00:00
* Please note that lk_exslpfail count may be lying
* about the real number of waiters with the
* LK_SLEEPFAIL flag on because they may be used in
* conjuction with interruptible sleeps so
Fix typos.
2010-01-07 01:24:09 +00:00
* lk_exslpfail might be considered an 'upper limit'
* bound, including the edge cases.
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
*/
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0);
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
realexslp = sleepq_sleepcnt(&lk->lock_object,
SQ_EXCLUSIVE_QUEUE);
if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) {
if (lk->lk_exslpfail < realexslp) {
lk->lk_exslpfail = 0;
queue = SQ_EXCLUSIVE_QUEUE;
v |= (x & LK_SHARED_WAITERS);
} else {
lk->lk_exslpfail = 0;
LOCK_LOG2(lk,
"%s: %p has only LK_SLEEPFAIL sleepers",
__func__, lk);
LOCK_LOG2(lk,
"%s: %p waking up threads on the exclusive queue",
__func__, lk);
wakeup_swapper =
sleepq_broadcast(&lk->lock_object,
SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE);
queue = SQ_SHARED_QUEUE;
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
} else {
Exclusive waiters sleeping with LK_SLEEPFAIL on and using interruptible sleeps/timeout may have left spourious lk_exslpfail counts on, so clean it up even when accessing a shared queue acquisition, giving to lk_exslpfail the value of 'upper limit'. In the worst case scenario, infact (mixed interruptible sleep / LK_SLEEPFAIL waiters) what may happen is that both queues are awaken even if that's not necessary, but still no harm. Reported by: Lucius Windschuh <lwindschuh at googlemail dot com> Reviewed by: kib Tested by: pho, Lucius Windschuh <lwindschuh at googlemail dot com>
2010-01-07 00:47:50 +00:00
/*
* Exclusive waiters sleeping with LK_SLEEPFAIL
* on and using interruptible sleeps/timeout
* may have left spourious lk_exslpfail counts
* on, so clean it up anyway.
*/
lk->lk_exslpfail = 0;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
queue = SQ_SHARED_QUEUE;
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
}
Print out a warning and a backtrace if we try to unlock a lockmgr that we do not hold. Glanced at by: phk MFC after: 3 days
2005-09-02 15:56:01 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
LOCK_LOG3(lk,
"%s: %p waking up threads on the %s queue",
__func__, lk, queue == SQ_SHARED_QUEUE ? "shared" :
"exclusive");
atomic_store_rel_ptr(&lk->lk_lock, v);
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
wakeup_swapper |= sleepq_broadcast(&lk->lock_object,
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
SLEEPQ_LK, 0, queue);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
sleepq_release(&lk->lock_object);
break;
} else
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
wakeup_swapper = wakeupshlk(lk, file, line);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
break;
case LK_DRAIN:
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
if (LK_CAN_WITNESS(flags))
WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER |
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
LOP_EXCLUSIVE, file, line, ilk);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
/*
s/alredy/already/ in the comments and the log message.
2008-07-25 11:22:25 +00:00
* Trying to drain a lock we already own will result in a
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
* deadlock.
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
*/
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (lockmgr_xlocked(lk)) {
if (flags & LK_INTERLOCK)
class->lc_unlock(ilk);
panic("%s: draining %s with the lock held @ %s:%d\n",
__func__, iwmesg, file, line);
}
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
while (!atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid)) {
Add software PMC support. New kernel events can be added at various location for sampling or counting. This will for example allow easy system profiling whatever the processor is with known tools like pmcstat(8). Simultaneous usage of software PMC and hardware PMC is possible, for example looking at the lock acquire failure, page fault while sampling on instructions. Sponsored by: NETASQ MFC after: 1 month
2012-03-28 20:58:30 +00:00
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lock_profile_obtain_lock_failed(&lk->lock_object,
&contested, &waittime);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* If the lock is expected to not sleep just give up
* and return.
*/
if (LK_TRYOP(flags)) {
LOCK_LOG2(lk, "%s: %p fails the try operation",
__func__, lk);
error = EBUSY;
break;
}
- Introduce lockmgr_args() in the lockmgr space. This function performs the same operation of lockmgr() but accepting a custom wmesg, prio and timo for the particular lock instance, overriding default values lkp->lk_wmesg, lkp->lk_prio and lkp->lk_timo. - Use lockmgr_args() in order to implement BUF_TIMELOCK() - Cleanup BUF_LOCK() - Remove LK_INTERNAL as it is nomore used in the lockmgr namespace Tested by: Andrea Barberio <insomniac at slackware dot it>
2008-02-15 21:04:36 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* Acquire the sleepqueue chain lock because we
* probabilly will need to manipulate waiters flags.
*/
sleepq_lock(&lk->lock_object);
x = lk->lk_lock;
Change the M_NAMEI allocations to use the zone allocator. This change plus the previous changes to use the zone allocator decrease the useage of malloc by half. The Zone allocator will be upgradeable to be able to use per CPU-pools, and has more intelligent usage of SPLs. Additionally, it has reasonable stats gathering capabilities, while making most calls inline.
1997-09-21 04:24:27 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* if the lock has been released while we spun on
* the sleepqueue chain lock just try again.
*/
if (x == LK_UNLOCKED) {
sleepq_release(&lk->lock_object);
continue;
}
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS);
if ((x & ~v) == LK_UNLOCKED) {
v = (x & ~LK_EXCLUSIVE_SPINNERS);
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
/*
* If interruptible sleeps left the exclusive
* queue empty avoid a starvation for the
* threads sleeping on the shared queue by
* giving them precedence and cleaning up the
* exclusive waiters bit anyway.
Tweak comments.
2010-01-07 01:19:01 +00:00
* Please note that lk_exslpfail count may be
* lying about the real number of waiters with
* the LK_SLEEPFAIL flag on because they may
* be used in conjuction with interruptible
Fix typos.
2010-01-07 01:24:09 +00:00
* sleeps so lk_exslpfail might be considered
* an 'upper limit' bound, including the edge
Tweak comments.
2010-01-07 01:19:01 +00:00
* cases.
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
*/
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (v & LK_EXCLUSIVE_WAITERS) {
queue = SQ_EXCLUSIVE_QUEUE;
v &= ~LK_EXCLUSIVE_WAITERS;
} else {
Exclusive waiters sleeping with LK_SLEEPFAIL on and using interruptible sleeps/timeout may have left spourious lk_exslpfail counts on, so clean it up even when accessing a shared queue acquisition, giving to lk_exslpfail the value of 'upper limit'. In the worst case scenario, infact (mixed interruptible sleep / LK_SLEEPFAIL waiters) what may happen is that both queues are awaken even if that's not necessary, but still no harm. Reported by: Lucius Windschuh <lwindschuh at googlemail dot com> Reviewed by: kib Tested by: pho, Lucius Windschuh <lwindschuh at googlemail dot com>
2010-01-07 00:47:50 +00:00
/*
* Exclusive waiters sleeping with
* LK_SLEEPFAIL on and using
* interruptible sleeps/timeout may
* have left spourious lk_exslpfail
* counts on, so clean it up anyway.
*/
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
MPASS(v & LK_SHARED_WAITERS);
Exclusive waiters sleeping with LK_SLEEPFAIL on and using interruptible sleeps/timeout may have left spourious lk_exslpfail counts on, so clean it up even when accessing a shared queue acquisition, giving to lk_exslpfail the value of 'upper limit'. In the worst case scenario, infact (mixed interruptible sleep / LK_SLEEPFAIL waiters) what may happen is that both queues are awaken even if that's not necessary, but still no harm. Reported by: Lucius Windschuh <lwindschuh at googlemail dot com> Reviewed by: kib Tested by: pho, Lucius Windschuh <lwindschuh at googlemail dot com>
2010-01-07 00:47:50 +00:00
lk->lk_exslpfail = 0;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
queue = SQ_SHARED_QUEUE;
v &= ~LK_SHARED_WAITERS;
}
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
if (queue == SQ_EXCLUSIVE_QUEUE) {
realexslp =
sleepq_sleepcnt(&lk->lock_object,
SQ_EXCLUSIVE_QUEUE);
if (lk->lk_exslpfail >= realexslp) {
lk->lk_exslpfail = 0;
queue = SQ_SHARED_QUEUE;
v &= ~LK_SHARED_WAITERS;
if (realexslp != 0) {
LOCK_LOG2(lk,
"%s: %p has only LK_SLEEPFAIL sleepers",
__func__, lk);
LOCK_LOG2(lk,
"%s: %p waking up threads on the exclusive queue",
__func__, lk);
wakeup_swapper =
sleepq_broadcast(
&lk->lock_object,
SLEEPQ_LK, 0,
SQ_EXCLUSIVE_QUEUE);
}
} else
lk->lk_exslpfail = 0;
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (!atomic_cmpset_ptr(&lk->lk_lock, x, v)) {
sleepq_release(&lk->lock_object);
continue;
}
LOCK_LOG3(lk,
"%s: %p waking up all threads on the %s queue",
__func__, lk, queue == SQ_SHARED_QUEUE ?
"shared" : "exclusive");
Use |= rather than += when aggregrating requests to wakeup the swapper. What we really want is an inclusive or of all the requests, and += can in theory roll over to 0.
2008-08-22 16:14:23 +00:00
wakeup_swapper |= sleepq_broadcast(
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
&lk->lock_object, SLEEPQ_LK, 0, queue);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* If shared waiters have been woken up we need
* to wait for one of them to acquire the lock
* before to set the exclusive waiters in
* order to avoid a deadlock.
*/
if (queue == SQ_SHARED_QUEUE) {
for (v = lk->lk_lock;
(v & LK_SHARE) && !LK_SHARERS(v);
v = lk->lk_lock)
cpu_spinwait();
}
}
/*
* Try to set the LK_EXCLUSIVE_WAITERS flag. If we
* fail, loop back and retry.
*/
if ((x & LK_EXCLUSIVE_WAITERS) == 0) {
if (!atomic_cmpset_ptr(&lk->lk_lock, x,
x | LK_EXCLUSIVE_WAITERS)) {
sleepq_release(&lk->lock_object);
continue;
}
LOCK_LOG2(lk, "%s: %p set drain waiters flag",
__func__, lk);
}
/*
* As far as we have been unable to acquire the
* exclusive lock and the exclusive waiters flag
* is set, we will sleep.
*/
if (flags & LK_INTERLOCK) {
class->lc_unlock(ilk);
flags &= ~LK_INTERLOCK;
}
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
GIANT_SAVE();
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
sleepq_add(&lk->lock_object, NULL, iwmesg, SLEEPQ_LK,
SQ_EXCLUSIVE_QUEUE);
sleepq_wait(&lk->lock_object, ipri & PRIMASK);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
GIANT_RESTORE();
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
LOCK_LOG2(lk, "%s: %p resuming from the sleep queue",
__func__, lk);
}
if (error == 0) {
lock_profile_obtain_lock_success(&lk->lock_object,
contested, waittime, file, line);
LOCK_LOG_LOCK("DRAIN", &lk->lock_object, 0,
lk->lk_recurse, file, line);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE |
LK_TRYWIT(flags), file, line);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
TD_LOCKS_INC(curthread);
STACK_SAVE(lk);
Change the M_NAMEI allocations to use the zone allocator. This change plus the previous changes to use the zone allocator decrease the useage of malloc by half. The Zone allocator will be upgradeable to be able to use per CPU-pools, and has more intelligent usage of SPLs. Additionally, it has reasonable stats gathering capabilities, while making most calls inline.
1997-09-21 04:24:27 +00:00
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
break;
default:
if (flags & LK_INTERLOCK)
class->lc_unlock(ilk);
panic("%s: unknown lockmgr request 0x%x\n", __func__, op);
Change the M_NAMEI allocations to use the zone allocator. This change plus the previous changes to use the zone allocator decrease the useage of malloc by half. The Zone allocator will be upgradeable to be able to use per CPU-pools, and has more intelligent usage of SPLs. Additionally, it has reasonable stats gathering capabilities, while making most calls inline.
1997-09-21 04:24:27 +00:00
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (flags & LK_INTERLOCK)
class->lc_unlock(ilk);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
if (wakeup_swapper)
kick_proc0();
Add WITNESS support to lockmgr locking primitive. This support tries to be as parallel as possible with other locking primitives, but there are differences; more specifically: - The base witness support is alredy equipped for allowing lock duplication acquisition as lockmgr rely on this. - In the case of lockmgr_disown() the lock result unlocked by witness even if it is still held by the "kernel context" - In the case of upgrading we can have 3 different situations: * Total unlocking of the shared lock and nothing else * Real witness upgrade if the owner is the first upgrader * Shared unlocking and exclusive locking if the owner is not the first upgrade but it is still allowed to upgrade - LK_DRAIN is basically handled like an exclusive acquisition Additively new options LK_NODUP and LK_NOWITNESS can now be used with lockinit(): LK_NOWITNESS disables WITNESS for the specified lock while LK_NODUP enable duplicated locks tracking. This will require manpages update and a __FreeBSD_version bumping (addressed by further commits). This patch also fixes a problem occurring if a lockmgr is held in exclusive mode and the same owner try to acquire it in shared mode: currently there is a spourious shared locking acquisition while what we really want is a lock downgrade. Probabilly, this situation can be better served with a EDEADLK failing errno return. Side note: first testing on this patch alredy reveleated several LORs reported, so please expect LORs cascades until resolved. NTFS also is reported broken by WITNESS introduction. BTW, NTFS is exposing a lock leak which needs to be fixed, and this patch can help it out if rightly tweaked. Tested by: kris, yar, Scot Hetzel <swhetzel at gmail dot com>
2008-02-06 00:37:14 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
return (error);
Change the M_NAMEI allocations to use the zone allocator. This change plus the previous changes to use the zone allocator decrease the useage of malloc by half. The Zone allocator will be upgradeable to be able to use per CPU-pools, and has more intelligent usage of SPLs. Additionally, it has reasonable stats gathering capabilities, while making most calls inline.
1997-09-21 04:24:27 +00:00
}
Convert lockmgr locks from using simple locks to using mutexes. Add lockdestroy() and appropriate invocations, which corresponds to lockinit() and must be called to clean up after a lockmgr lock is no longer needed.
2000-10-04 01:29:17 +00:00
void
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
_lockmgr_disown(struct lock *lk, const char *file, int line)
Convert lockmgr locks from using simple locks to using mutexes. Add lockdestroy() and appropriate invocations, which corresponds to lockinit() and must be called to clean up after a lockmgr lock is no longer needed.
2000-10-04 01:29:17 +00:00
{
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
uintptr_t tid, x;
panic: add a switch and infrastructure for stopping other CPUs in SMP case Historical behavior of letting other CPUs merily go on is a default for time being. The new behavior can be switched on via kern.stop_scheduler_on_panic tunable and sysctl. Stopping of the CPUs has (at least) the following benefits: - more of the system state at panic time is preserved intact - threads and interrupts do not interfere with dumping of the system state Only one thread runs uninterrupted after panic if stop_scheduler_on_panic is set. That thread might call code that is also used in normal context and that code might use locks to prevent concurrent execution of certain parts. Those locks might be held by the stopped threads and would never be released. To work around this issue, it was decided that instead of explicit checks for panic context, we would rather put those checks inside the locking primitives. This change has substantial portions written and re-written by attilio and kib at various times. Other changes are heavily based on the ideas and patches submitted by jhb and mdf. bde has provided many insights into the details and history of the current code. The new behavior may cause problems for systems that use a USB keyboard for interfacing with system console. This is because of some unusual locking patterns in the ukbd code which have to be used because on one hand ukbd is below syscons, but on the other hand it has to interface with other usb code that uses regular mutexes/Giant for its concurrency protection. Dumping to USB-connected disks may also be affected. PR: amd64/139614 (at least) In cooperation with: attilio, jhb, kib, mdf Discussed with: arch@, bde Tested by: Eugene Grosbein <eugen@grosbein.net>, gnn, Steven Hartland <killing@multiplay.co.uk>, glebius, Andrew Boyer <aboyer@averesystems.com> (various versions of the patch) MFC after: 3 months (or never)
2011-12-11 21:02:01 +00:00
if (SCHEDULER_STOPPED())
return;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
tid = (uintptr_t)curthread;
_lockmgr_assert(lk, KA_XLOCKED | KA_NOTRECURSED, file, line);
/*
s/alredy/already/ in the comments and the log message.
2008-07-25 11:22:25 +00:00
* If the owner is already LK_KERNPROC just skip the whole operation.
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
*/
if (LK_HOLDER(lk->lk_lock) != tid)
return;
- Call lock_profile_release when we're transitioning a lock to be owned by LK_KERNPROC. Discussed with: attilio
2009-03-14 11:43:02 +00:00
lock_profile_release_lock(&lk->lock_object);
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
LOCK_LOG_LOCK("XDISOWN", &lk->lock_object, 0, 0, file, line);
WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE, file, line);
TD_LOCKS_DEC(curthread);
Save the sack when doing a lockmgr_disown() call. Requested by: kib MFC: 3 days
2009-11-06 22:33:03 +00:00
STACK_SAVE(lk);
Move lock_profile_object_{init,destroy}() into lock_{init,destroy}().
2007-05-18 15:04:59 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
/*
* In order to preserve waiters flags, just spin.
*/
for (;;) {
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
x = lk->lk_lock;
MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0);
x &= LK_ALL_WAITERS;
Use a "rel" memory barrier for disowning the lock as it cames from an exclusive locking operation.
2008-04-13 01:21:56 +00:00
if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x,
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
LK_KERNPROC | x))
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
return;
cpu_spinwait();
}
Convert lockmgr locks from using simple locks to using mutexes. Add lockdestroy() and appropriate invocations, which corresponds to lockinit() and must be called to clean up after a lockmgr lock is no longer needed.
2000-10-04 01:29:17 +00:00
}
Remove explicit calling of lockmgr() with the NULL argument. Now, lockmgr() function can only be called passing curthread and the KASSERT() is upgraded according with this. In order to support on-the-fly owner switching, the new function lockmgr_disown() has been introduced and gets used in BUF_KERNPROC(). KPI, so, results changed and FreeBSD version will be bumped soon. Differently from previous code, we assume idle thread cannot try to acquire the lockmgr as it cannot sleep, so loose the relative check[1] in BUF_KERNPROC(). Tested by: kris [1] kib asked for a KASSERT in the lockmgr_disown() about this condition, but after thinking at it, as this is a well known general rule, I found it not really necessary.
2008-01-08 23:48:31 +00:00
void
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
lockmgr_printinfo(const struct lock *lk)
Remove explicit calling of lockmgr() with the NULL argument. Now, lockmgr() function can only be called passing curthread and the KASSERT() is upgraded according with this. In order to support on-the-fly owner switching, the new function lockmgr_disown() has been introduced and gets used in BUF_KERNPROC(). KPI, so, results changed and FreeBSD version will be bumped soon. Differently from previous code, we assume idle thread cannot try to acquire the lockmgr as it cannot sleep, so loose the relative check[1] in BUF_KERNPROC(). Tested by: kris [1] kib asked for a KASSERT in the lockmgr_disown() about this condition, but after thinking at it, as this is a well known general rule, I found it not really necessary.
2008-01-08 23:48:31 +00:00
{
struct thread *td;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
uintptr_t x;
if (lk->lk_lock == LK_UNLOCKED)
Tweak the output of VOP_PRINT/vn_printf() some. - Align the fifo output in fifo_print() with other vn_printf() output. - Remove the leading space from lockmgr_printinfo() so its output lines up in vn_printf(). - lockmgr_printinfo() now ends with a newline, so remove an extra newline from vn_printf().
2009-02-06 20:06:48 +00:00
printf("lock type %s: UNLOCKED\n", lk->lock_object.lo_name);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
else if (lk->lk_lock & LK_SHARE)
Tweak the output of VOP_PRINT/vn_printf() some. - Align the fifo output in fifo_print() with other vn_printf() output. - Remove the leading space from lockmgr_printinfo() so its output lines up in vn_printf(). - lockmgr_printinfo() now ends with a newline, so remove an extra newline from vn_printf().
2009-02-06 20:06:48 +00:00
printf("lock type %s: SHARED (count %ju)\n",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lock_object.lo_name,
(uintmax_t)LK_SHARERS(lk->lk_lock));
else {
td = lockmgr_xholder(lk);
Print out process name and thread id in the debugging message. This is useful because the message can end up in system logs in non-debugging operation. Reviewed by: attilio (earlier version)
2012-03-05 14:19:43 +00:00
printf("lock type %s: EXCL by thread %p "
"(pid %d, %s, tid %d)\n", lk->lock_object.lo_name, td,
td->td_proc->p_pid, td->td_proc->p_comm, td->td_tid);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
}
Remove explicit calling of lockmgr() with the NULL argument. Now, lockmgr() function can only be called passing curthread and the KASSERT() is upgraded according with this. In order to support on-the-fly owner switching, the new function lockmgr_disown() has been introduced and gets used in BUF_KERNPROC(). KPI, so, results changed and FreeBSD version will be bumped soon. Differently from previous code, we assume idle thread cannot try to acquire the lockmgr as it cannot sleep, so loose the relative check[1] in BUF_KERNPROC(). Tested by: kris [1] kib asked for a KASSERT in the lockmgr_disown() about this condition, but after thinking at it, as this is a well known general rule, I found it not really necessary.
2008-01-08 23:48:31 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
x = lk->lk_lock;
if (x & LK_EXCLUSIVE_WAITERS)
printf(" with exclusive waiters pending\n");
if (x & LK_SHARED_WAITERS)
printf(" with shared waiters pending\n");
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
if (x & LK_EXCLUSIVE_SPINNERS)
printf(" with exclusive spinners pending\n");
Remove explicit calling of lockmgr() with the NULL argument. Now, lockmgr() function can only be called passing curthread and the KASSERT() is upgraded according with this. In order to support on-the-fly owner switching, the new function lockmgr_disown() has been introduced and gets used in BUF_KERNPROC(). KPI, so, results changed and FreeBSD version will be bumped soon. Differently from previous code, we assume idle thread cannot try to acquire the lockmgr as it cannot sleep, so loose the relative check[1] in BUF_KERNPROC(). Tested by: kris [1] kib asked for a KASSERT in the lockmgr_disown() about this condition, but after thinking at it, as this is a well known general rule, I found it not really necessary.
2008-01-08 23:48:31 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
STACK_PRINT(lk);
Remove explicit calling of lockmgr() with the NULL argument. Now, lockmgr() function can only be called passing curthread and the KASSERT() is upgraded according with this. In order to support on-the-fly owner switching, the new function lockmgr_disown() has been introduced and gets used in BUF_KERNPROC(). KPI, so, results changed and FreeBSD version will be bumped soon. Differently from previous code, we assume idle thread cannot try to acquire the lockmgr as it cannot sleep, so loose the relative check[1] in BUF_KERNPROC(). Tested by: kris [1] kib asked for a KASSERT in the lockmgr_disown() about this condition, but after thinking at it, as this is a well known general rule, I found it not really necessary.
2008-01-08 23:48:31 +00:00
}
Change the M_NAMEI allocations to use the zone allocator. This change plus the previous changes to use the zone allocator decrease the useage of malloc by half. The Zone allocator will be upgradeable to be able to use per CPU-pools, and has more intelligent usage of SPLs. Additionally, it has reasonable stats gathering capabilities, while making most calls inline.
1997-09-21 04:24:27 +00:00
int
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
lockstatus(const struct lock *lk)
Change the M_NAMEI allocations to use the zone allocator. This change plus the previous changes to use the zone allocator decrease the useage of malloc by half. The Zone allocator will be upgradeable to be able to use per CPU-pools, and has more intelligent usage of SPLs. Additionally, it has reasonable stats gathering capabilities, while making most calls inline.
1997-09-21 04:24:27 +00:00
{
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
uintptr_t v, x;
int ret;
Change the M_NAMEI allocations to use the zone allocator. This change plus the previous changes to use the zone allocator decrease the useage of malloc by half. The Zone allocator will be upgradeable to be able to use per CPU-pools, and has more intelligent usage of SPLs. Additionally, it has reasonable stats gathering capabilities, while making most calls inline.
1997-09-21 04:24:27 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
ret = LK_SHARED;
x = lk->lk_lock;
v = LK_HOLDER(x);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if ((x & LK_SHARE) == 0) {
if (v == (uintptr_t)curthread || v == LK_KERNPROC)
ret = LK_EXCLUSIVE;
else
ret = LK_EXCLOTHER;
} else if (x == LK_UNLOCKED)
ret = 0;
return (ret);
Import 4.4BSD-Lite2 onto CSRG branch
1997-03-25 16:27:20 +00:00
}
Add a 'show lockmgr' command that dumps the relevant details of a lockmgr lock.
2006-08-15 16:42:16 +00:00
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
#ifdef INVARIANT_SUPPORT
Add some FEATURE macros for various features (AUDIT/CAM/IPC/KTR/MAC/NFS/NTP/ PMC/SYSV/...). No FreeBSD version bump, the userland application to query the features will be committed last and can serve as an indication of the availablility if needed. Sponsored by: Google Summer of Code 2010 Submitted by: kibab Reviewed by: arch@ (parts by rwatson, trasz, jhb) X-MFC after: to be determined in last commit with code from this project
2011-02-25 10:11:01 +00:00
FEATURE(invariant_support,
"Support for modules compiled with INVARIANTS option");
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
#ifndef INVARIANTS
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#undef _lockmgr_assert
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
#endif
void
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
_lockmgr_assert(const struct lock *lk, int what, const char *file, int line)
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
{
int slocked = 0;
if (panicstr != NULL)
return;
switch (what) {
case KA_SLOCKED:
case KA_SLOCKED | KA_NOTRECURSED:
case KA_SLOCKED | KA_RECURSED:
slocked = 1;
case KA_LOCKED:
case KA_LOCKED | KA_NOTRECURSED:
case KA_LOCKED | KA_RECURSED:
- Re-introduce WITNESS support for lockmgr. About the old implementation the only one difference is that lockmgr*() functions now accept LK_NOWITNESS flag which skips ordering for the instanced calling. - Remove an unuseful stub in witness_checkorder() (because the above check doesn't allow ever happening) and allow witness_upgrade() to accept non-try operation too.
2008-04-12 19:57:30 +00:00
#ifdef WITNESS
/*
* We cannot trust WITNESS if the lock is held in exclusive
* mode and a call to lockmgr_disown() happened.
* Workaround this skipping the check if the lock is held in
* exclusive mode even for the KA_LOCKED case.
*/
if (slocked || (lk->lk_lock & LK_SHARE)) {
witness_assert(&lk->lock_object, what, file, line);
break;
}
#endif
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (lk->lk_lock == LK_UNLOCKED ||
((lk->lk_lock & LK_SHARE) == 0 && (slocked ||
(!lockmgr_xlocked(lk) && !lockmgr_disowned(lk)))))
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
panic("Lock %s not %slocked @ %s:%d\n",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lock_object.lo_name, slocked ? "share" : "",
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
file, line);
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if ((lk->lk_lock & LK_SHARE) == 0) {
if (lockmgr_recursed(lk)) {
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
if (what & KA_NOTRECURSED)
panic("Lock %s recursed @ %s:%d\n",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lock_object.lo_name, file,
line);
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
} else if (what & KA_RECURSED)
panic("Lock %s not recursed @ %s:%d\n",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lock_object.lo_name, file, line);
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
}
break;
case KA_XLOCKED:
case KA_XLOCKED | KA_NOTRECURSED:
case KA_XLOCKED | KA_RECURSED:
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk))
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
panic("Lock %s not exclusively locked @ %s:%d\n",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lock_object.lo_name, file, line);
if (lockmgr_recursed(lk)) {
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
if (what & KA_NOTRECURSED)
panic("Lock %s recursed @ %s:%d\n",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lock_object.lo_name, file, line);
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
} else if (what & KA_RECURSED)
panic("Lock %s not recursed @ %s:%d\n",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lock_object.lo_name, file, line);
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
break;
case KA_UNLOCKED:
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (lockmgr_xlocked(lk) || lockmgr_disowned(lk))
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
panic("Lock %s exclusively locked @ %s:%d\n",
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk->lock_object.lo_name, file, line);
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
break;
default:
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
panic("Unknown lockmgr assertion: %d @ %s:%d\n", what, file,
line);
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
}
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
#endif
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
Add a 'show lockmgr' command that dumps the relevant details of a lockmgr lock.
2006-08-15 16:42:16 +00:00
#ifdef DDB
Add a new 'show sleepchain' ddb command similar to 'show lockchain' except that it operates on lockmgr and sx locks. This can be useful for tracking down vnode deadlocks in VFS for example. Note that this command is a bit more fragile than 'show lockchain' as we have to poke around at the wait channel of a thread to see if it points to either a struct lock or a condition variable inside of a struct sx. If td_wchan points to something unmapped, then this command will terminate early due to a fault, but no harm will be done.
2006-08-15 18:29:01 +00:00
int
lockmgr_chain(struct thread *td, struct thread **ownerp)
{
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
struct lock *lk;
Add a new 'show sleepchain' ddb command similar to 'show lockchain' except that it operates on lockmgr and sx locks. This can be useful for tracking down vnode deadlocks in VFS for example. Note that this command is a bit more fragile than 'show lockchain' as we have to poke around at the wait channel of a thread to see if it points to either a struct lock or a condition variable inside of a struct sx. If td_wchan points to something unmapped, then this command will terminate early due to a fault, but no harm will be done.
2006-08-15 18:29:01 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
lk = td->td_wchan;
Add a new 'show sleepchain' ddb command similar to 'show lockchain' except that it operates on lockmgr and sx locks. This can be useful for tracking down vnode deadlocks in VFS for example. Note that this command is a bit more fragile than 'show lockchain' as we have to poke around at the wait channel of a thread to see if it points to either a struct lock or a condition variable inside of a struct sx. If td_wchan points to something unmapped, then this command will terminate early due to a fault, but no harm will be done.
2006-08-15 18:29:01 +00:00
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (LOCK_CLASS(&lk->lock_object) != &lock_class_lockmgr)
return (0);
db_printf("blocked on lockmgr %s", lk->lock_object.lo_name);
if (lk->lk_lock & LK_SHARE)
db_printf("SHARED (count %ju)\n",
(uintmax_t)LK_SHARERS(lk->lk_lock));
else
db_printf("EXCL\n");
*ownerp = lockmgr_xholder(lk);
Handle the case when a thread is blocked on a lockmgr lock with LK_DRAIN in DDB's 'show sleepchain'. MFC after: 3 days
2007-03-21 19:28:20 +00:00
Add a new 'show sleepchain' ddb command similar to 'show lockchain' except that it operates on lockmgr and sx locks. This can be useful for tracking down vnode deadlocks in VFS for example. Note that this command is a bit more fragile than 'show lockchain' as we have to poke around at the wait channel of a thread to see if it points to either a struct lock or a condition variable inside of a struct sx. If td_wchan points to something unmapped, then this command will terminate early due to a fault, but no harm will be done.
2006-08-15 18:29:01 +00:00
return (1);
}
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
static void
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
db_show_lockmgr(const struct lock_object *lock)
Add a 'show lockmgr' command that dumps the relevant details of a lockmgr lock.
2006-08-15 16:42:16 +00:00
{
struct thread *td;
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
const struct lock *lk;
Add a 'show lockmgr' command that dumps the relevant details of a lockmgr lock.
2006-08-15 16:42:16 +00:00
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
lk = (const struct lock *)lock;
Add a 'show lockmgr' command that dumps the relevant details of a lockmgr lock.
2006-08-15 16:42:16 +00:00
- Use lock_init/lock_destroy() to setup the lock_object inside of lockmgr. We can now use LOCK_CLASS() as a stronger check in lockmgr_chain() as a result. This required putting back lk_flags as lockmgr's use of flags conflicted with other flags in lo_flags otherwise. - Tweak 'show lock' output for lockmgr to match sx, rw, and mtx.
2007-03-30 18:07:24 +00:00
db_printf(" state: ");
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
if (lk->lk_lock == LK_UNLOCKED)
Add a 'show lockmgr' command that dumps the relevant details of a lockmgr lock.
2006-08-15 16:42:16 +00:00
db_printf("UNLOCKED\n");
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
else if (lk->lk_lock & LK_SHARE)
db_printf("SLOCK: %ju\n", (uintmax_t)LK_SHARERS(lk->lk_lock));
else {
td = lockmgr_xholder(lk);
if (td == (struct thread *)LK_KERNPROC)
db_printf("XLOCK: LK_KERNPROC\n");
else
db_printf("XLOCK: %p (tid %d, pid %d, \"%s\")\n", td,
td->td_tid, td->td_proc->p_pid,
td->td_proc->p_comm);
if (lockmgr_recursed(lk))
db_printf(" recursed: %d\n", lk->lk_recurse);
}
db_printf(" waiters: ");
switch (lk->lk_lock & LK_ALL_WAITERS) {
case LK_SHARED_WAITERS:
db_printf("shared\n");
Add missing 'break' statement. Found with: Coverity Prevent(tm) CID: 3919
2009-05-12 17:05:40 +00:00
break;
Optimize lockmgr in order to get rid of the pool mutex interlock, of the state transitioning flags and of msleep(9) callings. Use, instead, an algorithm very similar to what sx(9) and rwlock(9) alredy do and direct accesses to the sleepqueue(9) primitive. In order to avoid writer starvation a mechanism very similar to what rwlock(9) uses now is implemented, with the correspective per-thread shared lockmgrs counter. This patch also adds 2 new functions to lockmgr KPI: lockmgr_rw() and lockmgr_args_rw(). These two are like the 2 "normal" versions, but they both accept a rwlock as interlock. In order to realize this, the general lockmgr manager function "__lockmgr_args()" has been implemented through the generic lock layer. It supports all the blocking primitives, but currently only these 2 mappers live. The patch drops the support for WITNESS atm, but it will be probabilly added soon. Also, there is a little race in the draining code which is also present in the current CVS stock implementation: if some sharers, once they wakeup, are in the runqueue they can contend the lock with the exclusive drainer. This is hard to be fixed but the now committed code mitigate this issue a lot better than the (past) CVS version. In addition assertive KA_HELD and KA_UNHELD have been made mute assertions because they are dangerous and they will be nomore supported soon. In order to avoid namespace pollution, stack.h is splitted into two parts: one which includes only the "struct stack" definition (_stack.h) and one defining the KPI. In this way, newly added _lockmgr.h can just include _stack.h. Kernel ABI results heavilly changed by this commit (the now committed version of "struct lock" is a lot smaller than the previous one) and KPI results broken by lockmgr_rw() / lockmgr_args_rw() introduction, so manpages and __FreeBSD_version will be updated accordingly. Tested by: kris, pho, jeff, danger Reviewed by: jeff Sponsored by: Google, Summer of Code program 2007
2008-04-06 20:08:51 +00:00
case LK_EXCLUSIVE_WAITERS:
db_printf("exclusive\n");
break;
case LK_ALL_WAITERS:
db_printf("shared and exclusive\n");
break;
default:
db_printf("none\n");
}
Introduce support for adaptive spinning in lockmgr. Actually, as it did receive few tuning, the support is disabled by default, but it can opt-in with the option ADAPTIVE_LOCKMGRS. Due to the nature of lockmgrs, adaptive spinning needs to be selectively enabled for any interested lockmgr. The support is bi-directional, or, in other ways, it will work in both cases if the lock is held in read or write way. In particular, the read path is passible of further tunning using the sysctls debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls should be axed or compiled out before release. Addictionally note that adaptive spinning doesn't cope well with LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers of LK_SLEEPFAIL are mainly interested in knowing if the interlock was dropped or not in order to reacquire it and re-test initial conditions. This directly interacts with adaptive spinning because lockmgr needs to drop the interlock while spinning in order to avoid a deadlock (further details in the comments inside the patch). Final note: finding someone willing to help on tuning this with relevant workloads would be either very important and appreciated. Tested by: jeff, pho Requested by: many
2009-06-17 01:55:42 +00:00
db_printf(" spinners: ");
if (lk->lk_lock & LK_EXCLUSIVE_SPINNERS)
db_printf("exclusive\n");
else
db_printf("none\n");
Add a 'show lockmgr' command that dumps the relevant details of a lockmgr lock.
2006-08-15 16:42:16 +00:00
}
#endif
Reference in New Issue Copy Permalink