1 .\" Copyright (c) 2007 Julian Elischer (julian - freebsd org )
2 .\" All rights reserved.
4 .\" Redistribution and use in source and binary forms, with or without
5 .\" modification, are permitted provided that the following conditions
7 .\" 1. Redistributions of source code must retain the above copyright
8 .\" notice, this list of conditions and the following disclaimer.
9 .\" 2. Redistributions in binary form must reproduce the above copyright
10 .\" notice, this list of conditions and the following disclaimer in the
11 .\" documentation and/or other materials provided with the distribution.
13 .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 .Nd kernel synchronization primitives
34 All sorts of stuff to go here.
39 kernel is written to run across multiple CPUs and as such requires
40 several different synchronization primitives to allow the developers
41 to safely access and manipulate the many data types required.
52 Shared-Exclusive locks
69 The primitives interact and have a number of rules regarding how
70 they can and can not be combined.
71 There are too many for the average
72 human mind and they keep changing.
73 (if you disagree, please write replacement text) :-)
75 Some of these primitives may be used at the low (interrupt) level and
78 There are strict ordering requirements and for some of the types this
84 Mutexes are the basic primitive.
85 You either hold it or you don't.
86 If you don't own it then you just spin, waiting for the holder (on
87 another CPU) to release it.
88 Hopefully they are doing something fast.
91 do anything that deschedules the thread while you
92 are holding a SPIN mutex.
94 Basically (regular) mutexes will deschedule the thread if the
95 mutex can not be acquired.
96 A non-spin mutex can be considered to be equivalent
97 to getting a write lock on an
99 (see below), and in fact non-spin mutexes and rw_locks may soon become the same thing.
100 As in spin mutexes, you either get it or you don't.
101 You may only call the
108 These will atomically drop the mutex and reacquire it
109 as part of waking up.
110 This is often however a
112 idea because it generally relies on you having
113 such a good knowledge of all the call graph above you
114 and what assumptions it is making that there are a lot
115 of ways to make hard-to-find mistakes.
116 For example you MUST re-test all the assumptions you made before,
117 all the way up the call graph to where you got the lock.
118 You can not just assume that mtx_sleep can be inserted anywhere.
119 If any caller above you has any mutex or
120 rwlock, your sleep, will cause a panic.
121 If the sleep only happens rarely it may be years before the
122 bad code path is found.
124 A variant of regular mutexes where the allocation of the mutex is handled
127 Reader/writer locks allow shared access to protected data by multiple threads,
128 or exclusive access by a single thread.
129 The threads with shared access are known as
131 since they should only read the protected data.
132 A thread with exclusive access is known as a
134 since it may modify protected data.
136 Although reader/writer locks look very similar to
138 (see below) locks, their usage pattern is different.
139 Reader/writer locks can be treated as mutexes (see above and
141 with shared/exclusive semantics.
142 More specifically, regular mutexes can be
143 considered to be equivalent to a write-lock on an
145 In the future this may in fact
146 become literally the fact.
149 can be locked while holding a regular mutex, but
152 be held while sleeping.
155 locks have priority propagation like mutexes, but priority
156 can be propagated only to an exclusive holder.
157 This limitation comes from the fact that shared owners
159 Another important property is that shared holders of
161 can recurse, but exclusive locks are not allowed to recurse.
162 This ability should not be used lightly and
164 Users of recursion in any locks should be prepared to
165 defend their decision against vigorous criticism.
167 Shared/exclusive locks are used to protect data that are read far more often
168 than they are written.
169 Mutexes are inherently more efficient than shared/exclusive locks, so
170 shared/exclusive locks should be used prudently.
171 The main reason for using an
173 is that a thread may hold a shared or exclusive lock on an
176 As a consequence of this however, an
178 lock may not be acquired while holding a mutex.
179 The reason for this is that, if one thread slept while holding an
181 lock while another thread blocked on the same
183 lock after acquiring a mutex, then the second thread would effectively
184 end up sleeping while holding a mutex, which is not allowed.
187 should be considered to be closely related to
189 In fact it could in some cases be
190 considered a conditional sleep.
192 Turnstiles are used to hold a queue of threads blocked on
194 Sleepable locks use condition variables to implement their queues.
195 Turnstiles differ from a sleep queue in that turnstile queue's
196 are assigned to a lock held by an owning thread.
197 Thus, when one thread is enqueued onto a turnstile, it can lend its
198 priority to the owning thread.
199 If this sounds confusing, we need to describe it better.
201 .Ss Condition variables
202 Condition variables are used in conjunction with mutexes to wait for
204 A thread must hold the mutex before calling the
207 When a thread waits on a condition, the mutex
208 is atomically released before the thread is blocked, then reacquired
209 before the function call returns.
211 Giant is a special instance of a sleep lock.
212 It has several special characteristics.
217 Drivers can request that Giant be locked around them, but this is
220 You can sleep while it has recursed, but other recursive locks cannot.
222 Giant must be locked first before other locks.
224 There are places in the kernel that drop Giant and pick it back up
226 Sleep locks will do this before sleeping.
227 Parts of the Network or VM code may do this as well, depending on the
229 This means that you cannot count on Giant keeping other code from
230 running if your code sleeps, even if you want it to.
241 handle event-based thread blocking.
242 If a thread must wait for an external event, it is put to sleep by
248 Threads may also wait using one of the locking primitive sleep routines
256 is an arbitrary address that uniquely identifies the event on which
257 the thread is being put to sleep.
258 All threads sleeping on a single
260 are woken up later by
262 often called from inside an interrupt routine, to indicate that the
263 resource the thread was blocking on is available now.
265 Several of the sleep functions including
268 and the locking primitive sleep routines specify an additional lock
270 The lock will be released before sleeping and reacquired
271 before the sleep routine returns.
276 flag, then the lock will not be reacquired before returning.
277 The lock is used to ensure that a condition can be checked atomically,
278 and that the current thread can be suspended without missing a
279 change to the condition, or an associated wakeup.
280 In addition, all of the sleep routines will fully drop the
284 while the thread is suspended and will reacquire the
286 mutex before the function returns.
288 .Ss lockmanager locks
292 page for more information.
293 I don't know what the downsides are but I'm sure someone will fill in this part.
295 .Ss Interaction table.
296 The following table shows what you can and can not do if you hold
297 one of the synchronization primitives discussed here:
298 (someone who knows what they are talking about should write this table)
299 .Bl -column ".Ic xxxxxxxxxxxxxxxxxxxx" ".Xr XXXXXXXXX" ".Xr XXXXXXX" ".Xr XXXXXXX" ".Xr XXXXXXX" ".Xr XXXXX" -offset indent
301 .Em "You have: You want:" Ta Spin_mtx Ta Slp_mtx Ta sx_lock Ta rw_lock Ta sleep
303 .It Ic SPIN mutex Ta \&ok-1 Ta \&no Ta \&no Ta \&no Ta \&no-3
304 .It Ic Sleep mutex Ta \&ok Ta \&ok-1 Ta \&no Ta \&ok Ta \&no-3
305 .It Ic sx_lock Ta \&ok Ta \&ok Ta \&ok-2 Ta \&ok Ta \&ok-4
306 .It Ic rw_lock Ta \&ok Ta \&ok Ta \&no Ta \&ok-2 Ta \&no-3
310 Recursion is defined per lock.
311 Lock order is important.
314 readers can recurse though writers can not.
315 Lock order is important.
318 There are calls atomically release this primitive when going to sleep
319 and reacquire it on wakeup (e.g.
327 Though one can sleep holding an sx lock, one can also use
329 which atomically release this primitive when going to sleep and
330 reacquire it on wakeup.
331 .Ss Context mode table.
332 The next table shows what can be used in different contexts.
333 At this time this is a rather easy to remember table.
334 .Bl -column ".Ic Xxxxxxxxxxxxxxxxxxxx" ".Xr XXXXXXXXX" ".Xr XXXXXXX" ".Xr XXXXXXX" ".Xr XXXXXXX" ".Xr XXXXX" -offset indent
336 .Em "Context:" Ta Spin_mtx Ta Slp_mtx Ta sx_lock Ta rw_lock Ta sleep
338 .It interrupt: Ta \&ok Ta \&no Ta \&no Ta \&no Ta \&no
339 .It idle: Ta \&ok Ta \&no Ta \&no Ta \&no Ta \&no
350 .Xr LOCK_PROFILING 9 ,
354 functions appeared in