xref: /freebsd/share/man/man9/locking.9 (revision 1669d8afc64812c8d2d1d147ae1fd42ff441e1b1)
1.\" Copyright (c) 2007 Julian Elischer  (julian -  freebsd org )
2.\" All rights reserved.
3.\"
4.\" Redistribution and use in source and binary forms, with or without
5.\" modification, are permitted provided that the following conditions
6.\" are met:
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.
12.\"
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
23.\" SUCH DAMAGE.
24.\"
25.\" $FreeBSD$
26.\"
27.Dd March 14, 2007
28.Dt LOCKING 9
29.Os
30.Sh NAME
31.Nm locking
32.Nd kernel synchronization primitives
33.Sh SYNOPSIS
34All sorts of stuff to go here.
35.Pp
36.Sh DESCRIPTION
37The
38.Em FreeBSD
39kernel is written to run across multiple CPUs and as such requires
40several different synchronization primitives to allow the developers
41to safely access and manipulate the many data types required.
42.Pp
43These include:
44.Bl -enum
45.It
46Spin Mutexes
47.It
48Sleep Mutexes
49.It
50pool Mutexes
51.It
52Shared-Exclusive locks
53.It
54Reader-Writer locks
55.It
56Read-Mostly locks
57.It
58Turnstiles
59.It
60Semaphores
61.It
62Condition variables
63.It
64Sleep/wakeup
65.It
66Giant
67.It
68Lockmanager locks
69.El
70.Pp
71The primitives interact and have a number of rules regarding how
72they can and can not be combined.
73There are too many for the average
74human mind and they keep changing.
75(if you disagree, please write replacement text)  :-)
76.Pp
77Some of these primitives may be used at the low (interrupt) level and
78some may not.
79.Pp
80There are strict ordering requirements and for some of the types this
81is checked using the
82.Xr witness 4
83code.
84.Pp
85.Ss SPIN Mutexes
86Mutexes are the basic primitive.
87You either hold it or you don't.
88If you don't own it then you just spin, waiting for the holder (on
89another CPU) to release it.
90Hopefully they are doing something fast.
91You
92.Em must not
93do anything that deschedules the thread while you
94are holding a SPIN mutex.
95.Ss Mutexes
96Basically (regular) mutexes will deschedule the thread if the
97mutex can not be acquired.
98A non-spin mutex can be considered to be equivalent
99to getting a write lock on an
100.Em rw_lock
101(see below), and in fact non-spin mutexes and rw_locks may soon become the same thing.
102As in spin mutexes, you either get it or you don't.
103You may only call the
104.Xr sleep 9
105call via
106.Fn msleep
107or the new
108.Fn mtx_sleep
109variant.
110These will atomically drop the mutex and reacquire it
111as part of waking up.
112This is often however a
113.Em BAD
114idea because it generally relies on you having
115such a good knowledge of all the call graph above you
116and what assumptions it is making that there are a lot
117of ways to make hard-to-find mistakes.
118For example you MUST re-test all the assumptions you made before,
119all the way up the call graph to where you got the lock.
120You can not just assume that mtx_sleep can be inserted anywhere.
121If any caller above you has any mutex or
122rwlock, your sleep, will cause a panic.
123If the sleep only happens rarely it may be years before the
124bad code path is found.
125.Ss Pool Mutexes
126A variant of regular mutexes where the allocation of the mutex is handled
127more by the system.
128.Ss Rw_locks
129Reader/writer locks allow shared access to protected data by multiple threads,
130or exclusive access by a single thread.
131The threads with shared access are known as
132.Em readers
133since they should only read the protected data.
134A thread with exclusive access is known as a
135.Em writer
136since it may modify protected data.
137.Pp
138Although reader/writer locks look very similar to
139.Xr sx 9
140(see below) locks, their usage pattern is different.
141Reader/writer locks can be treated as mutexes (see above and
142.Xr mutex 9 )
143with shared/exclusive semantics.
144More specifically, regular mutexes can be
145considered to be equivalent to a write-lock on an
146.Em rw_lock.
147In the future this may in fact
148become literally the fact.
149An
150.Em rw_lock
151can be locked while holding a regular mutex, but
152can
153.Em not
154be held while sleeping.
155The
156.Em rw_lock
157locks have priority propagation like mutexes, but priority
158can be propagated only to an exclusive holder.
159This limitation comes from the fact that shared owners
160are anonymous.
161Another important property is that shared holders of
162.Em rw_lock
163can recurse, but exclusive locks are not allowed to recurse.
164This ability should not be used lightly and
165.Em may go away.
166Users of recursion in any locks should be prepared to
167defend their decision against vigorous criticism.
168.Ss Rm_locks
169Mostly reader locks are similar to
170.Em Reader/write
171locks but optimized for very infrequent
172.Em writer
173locking.
174.Em rm_lock
175locks implement full priority propagation by tracking shared owners
176using a lock user supplied
177.Em tracker
178data structure.
179.Ss Sx_locks
180Shared/exclusive locks are used to protect data that are read far more often
181than they are written.
182Mutexes are inherently more efficient than shared/exclusive locks, so
183shared/exclusive locks should be used prudently.
184The main reason for using an
185.Em sx_lock
186is that a thread may hold a shared or exclusive lock on an
187.Em sx_lock
188lock while sleeping.
189As a consequence of this however, an
190.Em sx_lock
191lock may not be acquired while holding a mutex.
192The reason for this is that, if one thread slept while holding an
193.Em sx_lock
194lock while another thread blocked on the same
195.Em sx_lock
196lock after acquiring a mutex, then the second thread would effectively
197end up sleeping while holding a mutex, which is not allowed.
198The
199.Em sx_lock
200should be considered to be closely related to
201.Xr sleep 9 .
202In fact it could in some cases be
203considered a conditional sleep.
204.Ss Turnstiles
205Turnstiles are used to hold a queue of threads blocked on
206non-sleepable locks.
207Sleepable locks use condition variables to implement their queues.
208Turnstiles differ from a sleep queue in that turnstile queue's
209are assigned to a lock held by an owning thread.
210Thus, when one thread is enqueued onto a turnstile, it can lend its
211priority to the owning thread.
212If this sounds confusing, we need to describe it better.
213.Ss Semaphores
214.Ss Condition variables
215Condition variables are used in conjunction with mutexes to wait for
216conditions to occur.
217A thread must hold the mutex before calling the
218.Fn cv_wait* ,
219functions.
220When a thread waits on a condition, the mutex
221is atomically released before the thread is blocked, then reacquired
222before the function call returns.
223.Ss Giant
224Giant is a special instance of a sleep lock.
225It has several special characteristics.
226.Bl -enum
227.It
228It is recursive.
229.It
230Drivers can request that Giant be locked around them, but this is
231going away.
232.It
233You can sleep while it has recursed, but other recursive locks cannot.
234.It
235Giant must be locked first before other locks.
236.It
237There are places in the kernel that drop Giant and pick it back up
238again.
239Sleep locks will do this before sleeping.
240Parts of the Network or VM code may do this as well, depending on the
241setting of a sysctl.
242This means that you cannot count on Giant keeping other code from
243running if your code sleeps, even if you want it to.
244.El
245.Ss Sleep/wakeup
246The functions
247.Fn tsleep ,
248.Fn msleep ,
249.Fn msleep_spin ,
250.Fn pause ,
251.Fn wakeup ,
252and
253.Fn wakeup_one
254handle event-based thread blocking.
255If a thread must wait for an external event, it is put to sleep by
256.Fn tsleep ,
257.Fn msleep ,
258.Fn msleep_spin ,
259or
260.Fn pause .
261Threads may also wait using one of the locking primitive sleep routines
262.Xr mtx_sleep 9 ,
263.Xr rw_sleep 9 ,
264or
265.Xr sx_sleep 9 .
266.Pp
267The parameter
268.Fa chan
269is an arbitrary address that uniquely identifies the event on which
270the thread is being put to sleep.
271All threads sleeping on a single
272.Fa chan
273are woken up later by
274.Fn wakeup ,
275often called from inside an interrupt routine, to indicate that the
276resource the thread was blocking on is available now.
277.Pp
278Several of the sleep functions including
279.Fn msleep ,
280.Fn msleep_spin ,
281and the locking primitive sleep routines specify an additional lock
282parameter.
283The lock will be released before sleeping and reacquired
284before the sleep routine returns.
285If
286.Fa priority
287includes the
288.Dv PDROP
289flag, then the lock will not be reacquired before returning.
290The lock is used to ensure that a condition can be checked atomically,
291and that the current thread can be suspended without missing a
292change to the condition, or an associated wakeup.
293In addition, all of the sleep routines will fully drop the
294.Va Giant
295mutex
296(even if recursed)
297while the thread is suspended and will reacquire the
298.Va Giant
299mutex before the function returns.
300.Pp
301.Ss lockmanager locks
302Largely deprecated.
303See the
304.Xr lock 9
305page for more information.
306I don't know what the downsides are but I'm sure someone will fill in this part.
307.Sh Usage tables.
308.Ss Interaction table.
309The following table shows what you can and can not do if you hold
310one of the synchronization primitives discussed here:
311(someone who knows what they are talking about should write this table)
312.Bl -column ".Ic xxxxxxxxxxxxxxxxxxxx" ".Xr XXXXXXXXX" ".Xr XXXXXXX" ".Xr XXXXXXX" ".Xr XXXXXXX" ".Xr XXXXX" -offset indent
313.It Xo
314.Em "You have: You want:" Ta Spin_mtx Ta Slp_mtx Ta sx_lock Ta rw_lock Ta rm_lock Ta sleep
315.Xc
316.It Ic SPIN mutex  Ta \&ok-1 Ta \&no Ta \&no Ta \&no Ta \&no Ta \&no-3
317.It Ic Sleep mutex Ta \&ok Ta \&ok-1 Ta \&no Ta \&ok Ta \&ok Ta \&no-3
318.It Ic sx_lock     Ta \&ok Ta \&ok Ta \&ok-2 Ta \&ok Ta \&ok Ta \&ok-4
319.It Ic rw_lock     Ta \&ok Ta \&ok Ta \&no Ta \&ok-2 Ta \&ok Ta \&no-3
320.It Ic rm_lock     Ta \&ok Ta \&ok Ta \&no Ta \&ok Ta \&ok-2 Ta \&no
321.El
322.Pp
323.Em *1
324Recursion is defined per lock.
325Lock order is important.
326.Pp
327.Em *2
328readers can recurse though writers can not.
329Lock order is important.
330.Pp
331.Em *3
332There are calls atomically release this primitive when going to sleep
333and reacquire it on wakeup (e.g.
334.Fn mtx_sleep ,
335.Fn rw_sleep
336and
337.Fn msleep_spin
338).
339.Pp
340.Em *4
341Though one can sleep holding an sx lock, one can also use
342.Fn sx_sleep
343which atomically release this primitive when going to sleep and
344reacquire it on wakeup.
345.Ss Context mode table.
346The next table shows what can be used in different contexts.
347At this time this is a rather easy to remember table.
348.Bl -column ".Ic Xxxxxxxxxxxxxxxxxxxx" ".Xr XXXXXXXXX" ".Xr XXXXXXX" ".Xr XXXXXXX" ".Xr XXXXXXX" ".Xr XXXXX" -offset indent
349.It Xo
350.Em "Context:" Ta Spin_mtx Ta Slp_mtx Ta sx_lock Ta rw_lock Ta rm_lock Ta sleep
351.Xc
352.It interrupt:  Ta \&ok Ta \&no Ta \&no Ta \&no Ta \&no Ta \&no
353.It idle:  Ta \&ok Ta \&no Ta \&no Ta \&no Ta \&no Ta \&no
354.El
355.Sh SEE ALSO
356.Xr condvar 9 ,
357.Xr lock 9 ,
358.Xr mtx_pool 9 ,
359.Xr mutex 9 ,
360.Xr rmlock 9 ,
361.Xr rwlock 9 ,
362.Xr sema 9 ,
363.Xr sleep 9 ,
364.Xr sx 9 ,
365.Xr LOCK_PROFILING 9 ,
366.Xr WITNESS 9
367.Sh HISTORY
368These
369functions appeared in
370.Bsx 4.1
371through
372.Fx 7.0
373