1 .. _kernel_hacking_lock: 1 .. _kernel_hacking_lock:
2 2
3 =========================== 3 ===========================
4 Unreliable Guide To Locking 4 Unreliable Guide To Locking
5 =========================== 5 ===========================
6 6
7 :Author: Rusty Russell 7 :Author: Rusty Russell
8 8
9 Introduction 9 Introduction
10 ============ 10 ============
11 11
12 Welcome, to Rusty's Remarkably Unreliable Guid 12 Welcome, to Rusty's Remarkably Unreliable Guide to Kernel Locking
13 issues. This document describes the locking sy 13 issues. This document describes the locking systems in the Linux Kernel
14 in 2.6. 14 in 2.6.
15 15
16 With the wide availability of HyperThreading, 16 With the wide availability of HyperThreading, and preemption in the
17 Linux Kernel, everyone hacking on the kernel n 17 Linux Kernel, everyone hacking on the kernel needs to know the
18 fundamentals of concurrency and locking for SM 18 fundamentals of concurrency and locking for SMP.
19 19
20 The Problem With Concurrency 20 The Problem With Concurrency
21 ============================ 21 ============================
22 22
23 (Skip this if you know what a Race Condition i 23 (Skip this if you know what a Race Condition is).
24 24
25 In a normal program, you can increment a count 25 In a normal program, you can increment a counter like so:
26 26
27 :: 27 ::
28 28
29 very_important_count++; 29 very_important_count++;
30 30
31 31
32 This is what they would expect to happen: 32 This is what they would expect to happen:
33 33
34 34
35 .. table:: Expected Results 35 .. table:: Expected Results
36 36
37 +------------------------------------+------ 37 +------------------------------------+------------------------------------+
38 | Instance 1 | Insta 38 | Instance 1 | Instance 2 |
39 +====================================+====== 39 +====================================+====================================+
40 | read very_important_count (5) | 40 | read very_important_count (5) | |
41 +------------------------------------+------ 41 +------------------------------------+------------------------------------+
42 | add 1 (6) | 42 | add 1 (6) | |
43 +------------------------------------+------ 43 +------------------------------------+------------------------------------+
44 | write very_important_count (6) | 44 | write very_important_count (6) | |
45 +------------------------------------+------ 45 +------------------------------------+------------------------------------+
46 | | read 46 | | read very_important_count (6) |
47 +------------------------------------+------ 47 +------------------------------------+------------------------------------+
48 | | add 1 48 | | add 1 (7) |
49 +------------------------------------+------ 49 +------------------------------------+------------------------------------+
50 | | write 50 | | write very_important_count (7) |
51 +------------------------------------+------ 51 +------------------------------------+------------------------------------+
52 52
53 This is what might happen: 53 This is what might happen:
54 54
55 .. table:: Possible Results 55 .. table:: Possible Results
56 56
57 +------------------------------------+------ 57 +------------------------------------+------------------------------------+
58 | Instance 1 | Insta 58 | Instance 1 | Instance 2 |
59 +====================================+====== 59 +====================================+====================================+
60 | read very_important_count (5) | 60 | read very_important_count (5) | |
61 +------------------------------------+------ 61 +------------------------------------+------------------------------------+
62 | | read 62 | | read very_important_count (5) |
63 +------------------------------------+------ 63 +------------------------------------+------------------------------------+
64 | add 1 (6) | 64 | add 1 (6) | |
65 +------------------------------------+------ 65 +------------------------------------+------------------------------------+
66 | | add 1 66 | | add 1 (6) |
67 +------------------------------------+------ 67 +------------------------------------+------------------------------------+
68 | write very_important_count (6) | 68 | write very_important_count (6) | |
69 +------------------------------------+------ 69 +------------------------------------+------------------------------------+
70 | | write 70 | | write very_important_count (6) |
71 +------------------------------------+------ 71 +------------------------------------+------------------------------------+
72 72
73 73
74 Race Conditions and Critical Regions 74 Race Conditions and Critical Regions
75 ------------------------------------ 75 ------------------------------------
76 76
77 This overlap, where the result depends on the 77 This overlap, where the result depends on the relative timing of
78 multiple tasks, is called a race condition. Th 78 multiple tasks, is called a race condition. The piece of code containing
79 the concurrency issue is called a critical reg 79 the concurrency issue is called a critical region. And especially since
80 Linux starting running on SMP machines, they b 80 Linux starting running on SMP machines, they became one of the major
81 issues in kernel design and implementation. 81 issues in kernel design and implementation.
82 82
83 Preemption can have the same effect, even if t 83 Preemption can have the same effect, even if there is only one CPU: by
84 preempting one task during the critical region 84 preempting one task during the critical region, we have exactly the same
85 race condition. In this case the thread which 85 race condition. In this case the thread which preempts might run the
86 critical region itself. 86 critical region itself.
87 87
88 The solution is to recognize when these simult 88 The solution is to recognize when these simultaneous accesses occur, and
89 use locks to make sure that only one instance 89 use locks to make sure that only one instance can enter the critical
90 region at any time. There are many friendly pr 90 region at any time. There are many friendly primitives in the Linux
91 kernel to help you do this. And then there are 91 kernel to help you do this. And then there are the unfriendly
92 primitives, but I'll pretend they don't exist. 92 primitives, but I'll pretend they don't exist.
93 93
94 Locking in the Linux Kernel 94 Locking in the Linux Kernel
95 =========================== 95 ===========================
96 96
97 If I could give you one piece of advice on loc 97 If I could give you one piece of advice on locking: **keep it simple**.
98 98
99 Be reluctant to introduce new locks. 99 Be reluctant to introduce new locks.
100 100
101 Two Main Types of Kernel Locks: Spinlocks and 101 Two Main Types of Kernel Locks: Spinlocks and Mutexes
102 ---------------------------------------------- 102 -----------------------------------------------------
103 103
104 There are two main types of kernel locks. The 104 There are two main types of kernel locks. The fundamental type is the
105 spinlock (``include/asm/spinlock.h``), which i 105 spinlock (``include/asm/spinlock.h``), which is a very simple
106 single-holder lock: if you can't get the spinl 106 single-holder lock: if you can't get the spinlock, you keep trying
107 (spinning) until you can. Spinlocks are very s 107 (spinning) until you can. Spinlocks are very small and fast, and can be
108 used anywhere. 108 used anywhere.
109 109
110 The second type is a mutex (``include/linux/mu 110 The second type is a mutex (``include/linux/mutex.h``): it is like a
111 spinlock, but you may block holding a mutex. I 111 spinlock, but you may block holding a mutex. If you can't lock a mutex,
112 your task will suspend itself, and be woken up 112 your task will suspend itself, and be woken up when the mutex is
113 released. This means the CPU can do something 113 released. This means the CPU can do something else while you are
114 waiting. There are many cases when you simply 114 waiting. There are many cases when you simply can't sleep (see
115 `What Functions Are Safe To Call From Interrup 115 `What Functions Are Safe To Call From Interrupts?`_),
116 and so have to use a spinlock instead. 116 and so have to use a spinlock instead.
117 117
118 Neither type of lock is recursive: see 118 Neither type of lock is recursive: see
119 `Deadlock: Simple and Advanced`_. 119 `Deadlock: Simple and Advanced`_.
120 120
121 Locks and Uniprocessor Kernels 121 Locks and Uniprocessor Kernels
122 ------------------------------ 122 ------------------------------
123 123
124 For kernels compiled without ``CONFIG_SMP``, a 124 For kernels compiled without ``CONFIG_SMP``, and without
125 ``CONFIG_PREEMPT`` spinlocks do not exist at a 125 ``CONFIG_PREEMPT`` spinlocks do not exist at all. This is an excellent
126 design decision: when no-one else can run at t 126 design decision: when no-one else can run at the same time, there is no
127 reason to have a lock. 127 reason to have a lock.
128 128
129 If the kernel is compiled without ``CONFIG_SMP 129 If the kernel is compiled without ``CONFIG_SMP``, but ``CONFIG_PREEMPT``
130 is set, then spinlocks simply disable preempti 130 is set, then spinlocks simply disable preemption, which is sufficient to
131 prevent any races. For most purposes, we can t 131 prevent any races. For most purposes, we can think of preemption as
132 equivalent to SMP, and not worry about it sepa 132 equivalent to SMP, and not worry about it separately.
133 133
134 You should always test your locking code with 134 You should always test your locking code with ``CONFIG_SMP`` and
135 ``CONFIG_PREEMPT`` enabled, even if you don't 135 ``CONFIG_PREEMPT`` enabled, even if you don't have an SMP test box,
136 because it will still catch some kinds of lock 136 because it will still catch some kinds of locking bugs.
137 137
138 Mutexes still exist, because they are required 138 Mutexes still exist, because they are required for synchronization
139 between user contexts, as we will see below. 139 between user contexts, as we will see below.
140 140
141 Locking Only In User Context 141 Locking Only In User Context
142 ---------------------------- 142 ----------------------------
143 143
144 If you have a data structure which is only eve 144 If you have a data structure which is only ever accessed from user
145 context, then you can use a simple mutex (``in 145 context, then you can use a simple mutex (``include/linux/mutex.h``) to
146 protect it. This is the most trivial case: you 146 protect it. This is the most trivial case: you initialize the mutex.
147 Then you can call mutex_lock_interruptible() t 147 Then you can call mutex_lock_interruptible() to grab the
148 mutex, and mutex_unlock() to release it. There 148 mutex, and mutex_unlock() to release it. There is also a
149 mutex_lock(), which should be avoided, because 149 mutex_lock(), which should be avoided, because it will
150 not return if a signal is received. 150 not return if a signal is received.
151 151
152 Example: ``net/netfilter/nf_sockopt.c`` allows 152 Example: ``net/netfilter/nf_sockopt.c`` allows registration of new
153 setsockopt() and getsockopt() calls, with 153 setsockopt() and getsockopt() calls, with
154 nf_register_sockopt(). Registration and de-reg 154 nf_register_sockopt(). Registration and de-registration
155 are only done on module load and unload (and b 155 are only done on module load and unload (and boot time, where there is
156 no concurrency), and the list of registrations 156 no concurrency), and the list of registrations is only consulted for an
157 unknown setsockopt() or getsockopt() system 157 unknown setsockopt() or getsockopt() system
158 call. The ``nf_sockopt_mutex`` is perfect to p 158 call. The ``nf_sockopt_mutex`` is perfect to protect this, especially
159 since the setsockopt and getsockopt calls may 159 since the setsockopt and getsockopt calls may well sleep.
160 160
161 Locking Between User Context and Softirqs 161 Locking Between User Context and Softirqs
162 ----------------------------------------- 162 -----------------------------------------
163 163
164 If a softirq shares data with user context, yo 164 If a softirq shares data with user context, you have two problems.
165 Firstly, the current user context can be inter 165 Firstly, the current user context can be interrupted by a softirq, and
166 secondly, the critical region could be entered 166 secondly, the critical region could be entered from another CPU. This is
167 where spin_lock_bh() (``include/linux/spinlock 167 where spin_lock_bh() (``include/linux/spinlock.h``) is
168 used. It disables softirqs on that CPU, then g 168 used. It disables softirqs on that CPU, then grabs the lock.
169 spin_unlock_bh() does the reverse. (The '_bh' 169 spin_unlock_bh() does the reverse. (The '_bh' suffix is
170 a historical reference to "Bottom Halves", the 170 a historical reference to "Bottom Halves", the old name for software
171 interrupts. It should really be called spin_lo 171 interrupts. It should really be called spin_lock_softirq()' in a
172 perfect world). 172 perfect world).
173 173
174 Note that you can also use spin_lock_irq() or 174 Note that you can also use spin_lock_irq() or
175 spin_lock_irqsave() here, which stop hardware 175 spin_lock_irqsave() here, which stop hardware interrupts
176 as well: see `Hard IRQ Context`_. 176 as well: see `Hard IRQ Context`_.
177 177
178 This works perfectly for UP as well: the spin 178 This works perfectly for UP as well: the spin lock vanishes, and this
179 macro simply becomes local_bh_disable() 179 macro simply becomes local_bh_disable()
180 (``include/linux/interrupt.h``), which protect 180 (``include/linux/interrupt.h``), which protects you from the softirq
181 being run. 181 being run.
182 182
183 Locking Between User Context and Tasklets 183 Locking Between User Context and Tasklets
184 ----------------------------------------- 184 -----------------------------------------
185 185
186 This is exactly the same as above, because tas 186 This is exactly the same as above, because tasklets are actually run
187 from a softirq. 187 from a softirq.
188 188
189 Locking Between User Context and Timers 189 Locking Between User Context and Timers
190 --------------------------------------- 190 ---------------------------------------
191 191
192 This, too, is exactly the same as above, becau 192 This, too, is exactly the same as above, because timers are actually run
193 from a softirq. From a locking point of view, 193 from a softirq. From a locking point of view, tasklets and timers are
194 identical. 194 identical.
195 195
196 Locking Between Tasklets/Timers 196 Locking Between Tasklets/Timers
197 ------------------------------- 197 -------------------------------
198 198
199 Sometimes a tasklet or timer might want to sha 199 Sometimes a tasklet or timer might want to share data with another
200 tasklet or timer. 200 tasklet or timer.
201 201
202 The Same Tasklet/Timer 202 The Same Tasklet/Timer
203 ~~~~~~~~~~~~~~~~~~~~~~ 203 ~~~~~~~~~~~~~~~~~~~~~~
204 204
205 Since a tasklet is never run on two CPUs at on 205 Since a tasklet is never run on two CPUs at once, you don't need to
206 worry about your tasklet being reentrant (runn 206 worry about your tasklet being reentrant (running twice at once), even
207 on SMP. 207 on SMP.
208 208
209 Different Tasklets/Timers 209 Different Tasklets/Timers
210 ~~~~~~~~~~~~~~~~~~~~~~~~~ 210 ~~~~~~~~~~~~~~~~~~~~~~~~~
211 211
212 If another tasklet/timer wants to share data w 212 If another tasklet/timer wants to share data with your tasklet or timer
213 , you will both need to use spin_lock() and 213 , you will both need to use spin_lock() and
214 spin_unlock() calls. spin_lock_bh() is 214 spin_unlock() calls. spin_lock_bh() is
215 unnecessary here, as you are already in a task 215 unnecessary here, as you are already in a tasklet, and none will be run
216 on the same CPU. 216 on the same CPU.
217 217
218 Locking Between Softirqs 218 Locking Between Softirqs
219 ------------------------ 219 ------------------------
220 220
221 Often a softirq might want to share data with 221 Often a softirq might want to share data with itself or a tasklet/timer.
222 222
223 The Same Softirq 223 The Same Softirq
224 ~~~~~~~~~~~~~~~~ 224 ~~~~~~~~~~~~~~~~
225 225
226 The same softirq can run on the other CPUs: yo 226 The same softirq can run on the other CPUs: you can use a per-CPU array
227 (see `Per-CPU Data`_) for better performance. 227 (see `Per-CPU Data`_) for better performance. If you're
228 going so far as to use a softirq, you probably 228 going so far as to use a softirq, you probably care about scalable
229 performance enough to justify the extra comple 229 performance enough to justify the extra complexity.
230 230
231 You'll need to use spin_lock() and 231 You'll need to use spin_lock() and
232 spin_unlock() for shared data. 232 spin_unlock() for shared data.
233 233
234 Different Softirqs 234 Different Softirqs
235 ~~~~~~~~~~~~~~~~~~ 235 ~~~~~~~~~~~~~~~~~~
236 236
237 You'll need to use spin_lock() and 237 You'll need to use spin_lock() and
238 spin_unlock() for shared data, whether it be a 238 spin_unlock() for shared data, whether it be a timer,
239 tasklet, different softirq or the same or anot 239 tasklet, different softirq or the same or another softirq: any of them
240 could be running on a different CPU. 240 could be running on a different CPU.
241 241
242 Hard IRQ Context 242 Hard IRQ Context
243 ================ 243 ================
244 244
245 Hardware interrupts usually communicate with a 245 Hardware interrupts usually communicate with a tasklet or softirq.
246 Frequently this involves putting work in a que 246 Frequently this involves putting work in a queue, which the softirq will
247 take out. 247 take out.
248 248
249 Locking Between Hard IRQ and Softirqs/Tasklets 249 Locking Between Hard IRQ and Softirqs/Tasklets
250 ---------------------------------------------- 250 ----------------------------------------------
251 251
252 If a hardware irq handler shares data with a s 252 If a hardware irq handler shares data with a softirq, you have two
253 concerns. Firstly, the softirq processing can 253 concerns. Firstly, the softirq processing can be interrupted by a
254 hardware interrupt, and secondly, the critical 254 hardware interrupt, and secondly, the critical region could be entered
255 by a hardware interrupt on another CPU. This i 255 by a hardware interrupt on another CPU. This is where
256 spin_lock_irq() is used. It is defined to disa 256 spin_lock_irq() is used. It is defined to disable
257 interrupts on that cpu, then grab the lock. 257 interrupts on that cpu, then grab the lock.
258 spin_unlock_irq() does the reverse. 258 spin_unlock_irq() does the reverse.
259 259
260 The irq handler does not need to use spin_lock 260 The irq handler does not need to use spin_lock_irq(), because
261 the softirq cannot run while the irq handler i 261 the softirq cannot run while the irq handler is running: it can use
262 spin_lock(), which is slightly faster. The onl 262 spin_lock(), which is slightly faster. The only exception
263 would be if a different hardware irq handler u 263 would be if a different hardware irq handler uses the same lock:
264 spin_lock_irq() will stop that from interrupti 264 spin_lock_irq() will stop that from interrupting us.
265 265
266 This works perfectly for UP as well: the spin 266 This works perfectly for UP as well: the spin lock vanishes, and this
267 macro simply becomes local_irq_disable() 267 macro simply becomes local_irq_disable()
268 (``include/asm/smp.h``), which protects you fr 268 (``include/asm/smp.h``), which protects you from the softirq/tasklet/BH
269 being run. 269 being run.
270 270
271 spin_lock_irqsave() (``include/linux/spinlock. 271 spin_lock_irqsave() (``include/linux/spinlock.h``) is a
272 variant which saves whether interrupts were on 272 variant which saves whether interrupts were on or off in a flags word,
273 which is passed to spin_unlock_irqrestore(). T 273 which is passed to spin_unlock_irqrestore(). This means
274 that the same code can be used inside an hard 274 that the same code can be used inside an hard irq handler (where
275 interrupts are already off) and in softirqs (w 275 interrupts are already off) and in softirqs (where the irq disabling is
276 required). 276 required).
277 277
278 Note that softirqs (and hence tasklets and tim 278 Note that softirqs (and hence tasklets and timers) are run on return
279 from hardware interrupts, so spin_lock_irq() a 279 from hardware interrupts, so spin_lock_irq() also stops
280 these. In that sense, spin_lock_irqsave() is t 280 these. In that sense, spin_lock_irqsave() is the most
281 general and powerful locking function. 281 general and powerful locking function.
282 282
283 Locking Between Two Hard IRQ Handlers 283 Locking Between Two Hard IRQ Handlers
284 ------------------------------------- 284 -------------------------------------
285 285
286 It is rare to have to share data between two I 286 It is rare to have to share data between two IRQ handlers, but if you
287 do, spin_lock_irqsave() should be used: it is 287 do, spin_lock_irqsave() should be used: it is
288 architecture-specific whether all interrupts a 288 architecture-specific whether all interrupts are disabled inside irq
289 handlers themselves. 289 handlers themselves.
290 290
291 Cheat Sheet For Locking 291 Cheat Sheet For Locking
292 ======================= 292 =======================
293 293
294 Pete Zaitcev gives the following summary: 294 Pete Zaitcev gives the following summary:
295 295
296 - If you are in a process context (any syscal 296 - If you are in a process context (any syscall) and want to lock other
297 process out, use a mutex. You can take a mu 297 process out, use a mutex. You can take a mutex and sleep
298 (``copy_from_user()`` or ``kmalloc(x,GFP_KE !! 298 (``copy_from_user*(`` or ``kmalloc(x,GFP_KERNEL)``).
299 299
300 - Otherwise (== data can be touched in an int 300 - Otherwise (== data can be touched in an interrupt), use
301 spin_lock_irqsave() and 301 spin_lock_irqsave() and
302 spin_unlock_irqrestore(). 302 spin_unlock_irqrestore().
303 303
304 - Avoid holding spinlock for more than 5 line 304 - Avoid holding spinlock for more than 5 lines of code and across any
305 function call (except accessors like readb( 305 function call (except accessors like readb()).
306 306
307 Table of Minimum Requirements 307 Table of Minimum Requirements
308 ----------------------------- 308 -----------------------------
309 309
310 The following table lists the **minimum** lock 310 The following table lists the **minimum** locking requirements between
311 various contexts. In some cases, the same cont 311 various contexts. In some cases, the same context can only be running on
312 one CPU at a time, so no locking is required f 312 one CPU at a time, so no locking is required for that context (eg. a
313 particular thread can only run on one CPU at a 313 particular thread can only run on one CPU at a time, but if it needs
314 shares data with another thread, locking is re 314 shares data with another thread, locking is required).
315 315
316 Remember the advice above: you can always use 316 Remember the advice above: you can always use
317 spin_lock_irqsave(), which is a superset of al 317 spin_lock_irqsave(), which is a superset of all other
318 spinlock primitives. 318 spinlock primitives.
319 319
320 ============== ============= ============= === 320 ============== ============= ============= ========= ========= ========= ========= ======= ======= ============== ==============
321 . IRQ Handler A IRQ Handler B Sof 321 . IRQ Handler A IRQ Handler B Softirq A Softirq B Tasklet A Tasklet B Timer A Timer B User Context A User Context B
322 ============== ============= ============= === 322 ============== ============= ============= ========= ========= ========= ========= ======= ======= ============== ==============
323 IRQ Handler A None 323 IRQ Handler A None
324 IRQ Handler B SLIS None 324 IRQ Handler B SLIS None
325 Softirq A SLI SLI SL 325 Softirq A SLI SLI SL
326 Softirq B SLI SLI SL 326 Softirq B SLI SLI SL SL
327 Tasklet A SLI SLI SL 327 Tasklet A SLI SLI SL SL None
328 Tasklet B SLI SLI SL 328 Tasklet B SLI SLI SL SL SL None
329 Timer A SLI SLI SL 329 Timer A SLI SLI SL SL SL SL None
330 Timer B SLI SLI SL 330 Timer B SLI SLI SL SL SL SL SL None
331 User Context A SLI SLI SLB 331 User Context A SLI SLI SLBH SLBH SLBH SLBH SLBH SLBH None
332 User Context B SLI SLI SLB 332 User Context B SLI SLI SLBH SLBH SLBH SLBH SLBH SLBH MLI None
333 ============== ============= ============= === 333 ============== ============= ============= ========= ========= ========= ========= ======= ======= ============== ==============
334 334
335 Table: Table of Locking Requirements 335 Table: Table of Locking Requirements
336 336
337 +--------+----------------------------+ 337 +--------+----------------------------+
338 | SLIS | spin_lock_irqsave | 338 | SLIS | spin_lock_irqsave |
339 +--------+----------------------------+ 339 +--------+----------------------------+
340 | SLI | spin_lock_irq | 340 | SLI | spin_lock_irq |
341 +--------+----------------------------+ 341 +--------+----------------------------+
342 | SL | spin_lock | 342 | SL | spin_lock |
343 +--------+----------------------------+ 343 +--------+----------------------------+
344 | SLBH | spin_lock_bh | 344 | SLBH | spin_lock_bh |
345 +--------+----------------------------+ 345 +--------+----------------------------+
346 | MLI | mutex_lock_interruptible | 346 | MLI | mutex_lock_interruptible |
347 +--------+----------------------------+ 347 +--------+----------------------------+
348 348
349 Table: Legend for Locking Requirements Table 349 Table: Legend for Locking Requirements Table
350 350
351 The trylock Functions 351 The trylock Functions
352 ===================== 352 =====================
353 353
354 There are functions that try to acquire a lock 354 There are functions that try to acquire a lock only once and immediately
355 return a value telling about success or failur 355 return a value telling about success or failure to acquire the lock.
356 They can be used if you need no access to the 356 They can be used if you need no access to the data protected with the
357 lock when some other thread is holding the loc 357 lock when some other thread is holding the lock. You should acquire the
358 lock later if you then need access to the data 358 lock later if you then need access to the data protected with the lock.
359 359
360 spin_trylock() does not spin but returns non-z 360 spin_trylock() does not spin but returns non-zero if it
361 acquires the spinlock on the first try or 0 if 361 acquires the spinlock on the first try or 0 if not. This function can be
362 used in all contexts like spin_lock(): you mus 362 used in all contexts like spin_lock(): you must have
363 disabled the contexts that might interrupt you 363 disabled the contexts that might interrupt you and acquire the spin
364 lock. 364 lock.
365 365
366 mutex_trylock() does not suspend your task but 366 mutex_trylock() does not suspend your task but returns
367 non-zero if it could lock the mutex on the fir 367 non-zero if it could lock the mutex on the first try or 0 if not. This
368 function cannot be safely used in hardware or 368 function cannot be safely used in hardware or software interrupt
369 contexts despite not sleeping. 369 contexts despite not sleeping.
370 370
371 Common Examples 371 Common Examples
372 =============== 372 ===============
373 373
374 Let's step through a simple example: a cache o 374 Let's step through a simple example: a cache of number to name mappings.
375 The cache keeps a count of how often each of t 375 The cache keeps a count of how often each of the objects is used, and
376 when it gets full, throws out the least used o 376 when it gets full, throws out the least used one.
377 377
378 All In User Context 378 All In User Context
379 ------------------- 379 -------------------
380 380
381 For our first example, we assume that all oper 381 For our first example, we assume that all operations are in user context
382 (ie. from system calls), so we can sleep. This 382 (ie. from system calls), so we can sleep. This means we can use a mutex
383 to protect the cache and all the objects withi 383 to protect the cache and all the objects within it. Here's the code::
384 384
385 #include <linux/list.h> 385 #include <linux/list.h>
386 #include <linux/slab.h> 386 #include <linux/slab.h>
387 #include <linux/string.h> 387 #include <linux/string.h>
388 #include <linux/mutex.h> 388 #include <linux/mutex.h>
389 #include <asm/errno.h> 389 #include <asm/errno.h>
390 390
391 struct object 391 struct object
392 { 392 {
393 struct list_head list; 393 struct list_head list;
394 int id; 394 int id;
395 char name[32]; 395 char name[32];
396 int popularity; 396 int popularity;
397 }; 397 };
398 398
399 /* Protects the cache, cache_num, and the 399 /* Protects the cache, cache_num, and the objects within it */
400 static DEFINE_MUTEX(cache_lock); 400 static DEFINE_MUTEX(cache_lock);
401 static LIST_HEAD(cache); 401 static LIST_HEAD(cache);
402 static unsigned int cache_num = 0; 402 static unsigned int cache_num = 0;
403 #define MAX_CACHE_SIZE 10 403 #define MAX_CACHE_SIZE 10
404 404
405 /* Must be holding cache_lock */ 405 /* Must be holding cache_lock */
406 static struct object *__cache_find(int id) 406 static struct object *__cache_find(int id)
407 { 407 {
408 struct object *i; 408 struct object *i;
409 409
410 list_for_each_entry(i, &cache, lis 410 list_for_each_entry(i, &cache, list)
411 if (i->id == id) { 411 if (i->id == id) {
412 i->popularity++; 412 i->popularity++;
413 return i; 413 return i;
414 } 414 }
415 return NULL; 415 return NULL;
416 } 416 }
417 417
418 /* Must be holding cache_lock */ 418 /* Must be holding cache_lock */
419 static void __cache_delete(struct object * 419 static void __cache_delete(struct object *obj)
420 { 420 {
421 BUG_ON(!obj); 421 BUG_ON(!obj);
422 list_del(&obj->list); 422 list_del(&obj->list);
423 kfree(obj); 423 kfree(obj);
424 cache_num--; 424 cache_num--;
425 } 425 }
426 426
427 /* Must be holding cache_lock */ 427 /* Must be holding cache_lock */
428 static void __cache_add(struct object *obj 428 static void __cache_add(struct object *obj)
429 { 429 {
430 list_add(&obj->list, &cache); 430 list_add(&obj->list, &cache);
431 if (++cache_num > MAX_CACHE_SIZE) 431 if (++cache_num > MAX_CACHE_SIZE) {
432 struct object *i, *outcast 432 struct object *i, *outcast = NULL;
433 list_for_each_entry(i, &ca 433 list_for_each_entry(i, &cache, list) {
434 if (!outcast || i- 434 if (!outcast || i->popularity < outcast->popularity)
435 outcast = 435 outcast = i;
436 } 436 }
437 __cache_delete(outcast); 437 __cache_delete(outcast);
438 } 438 }
439 } 439 }
440 440
441 int cache_add(int id, const char *name) 441 int cache_add(int id, const char *name)
442 { 442 {
443 struct object *obj; 443 struct object *obj;
444 444
445 if ((obj = kmalloc(sizeof(*obj), G 445 if ((obj = kmalloc(sizeof(*obj), GFP_KERNEL)) == NULL)
446 return -ENOMEM; 446 return -ENOMEM;
447 447
448 strscpy(obj->name, name, sizeof(ob 448 strscpy(obj->name, name, sizeof(obj->name));
449 obj->id = id; 449 obj->id = id;
450 obj->popularity = 0; 450 obj->popularity = 0;
451 451
452 mutex_lock(&cache_lock); 452 mutex_lock(&cache_lock);
453 __cache_add(obj); 453 __cache_add(obj);
454 mutex_unlock(&cache_lock); 454 mutex_unlock(&cache_lock);
455 return 0; 455 return 0;
456 } 456 }
457 457
458 void cache_delete(int id) 458 void cache_delete(int id)
459 { 459 {
460 mutex_lock(&cache_lock); 460 mutex_lock(&cache_lock);
461 __cache_delete(__cache_find(id)); 461 __cache_delete(__cache_find(id));
462 mutex_unlock(&cache_lock); 462 mutex_unlock(&cache_lock);
463 } 463 }
464 464
465 int cache_find(int id, char *name) 465 int cache_find(int id, char *name)
466 { 466 {
467 struct object *obj; 467 struct object *obj;
468 int ret = -ENOENT; 468 int ret = -ENOENT;
469 469
470 mutex_lock(&cache_lock); 470 mutex_lock(&cache_lock);
471 obj = __cache_find(id); 471 obj = __cache_find(id);
472 if (obj) { 472 if (obj) {
473 ret = 0; 473 ret = 0;
474 strcpy(name, obj->name); 474 strcpy(name, obj->name);
475 } 475 }
476 mutex_unlock(&cache_lock); 476 mutex_unlock(&cache_lock);
477 return ret; 477 return ret;
478 } 478 }
479 479
480 Note that we always make sure we have the cach 480 Note that we always make sure we have the cache_lock when we add,
481 delete, or look up the cache: both the cache i 481 delete, or look up the cache: both the cache infrastructure itself and
482 the contents of the objects are protected by t 482 the contents of the objects are protected by the lock. In this case it's
483 easy, since we copy the data for the user, and 483 easy, since we copy the data for the user, and never let them access the
484 objects directly. 484 objects directly.
485 485
486 There is a slight (and common) optimization he 486 There is a slight (and common) optimization here: in
487 cache_add() we set up the fields of the object 487 cache_add() we set up the fields of the object before
488 grabbing the lock. This is safe, as no-one els 488 grabbing the lock. This is safe, as no-one else can access it until we
489 put it in cache. 489 put it in cache.
490 490
491 Accessing From Interrupt Context 491 Accessing From Interrupt Context
492 -------------------------------- 492 --------------------------------
493 493
494 Now consider the case where cache_find() can b 494 Now consider the case where cache_find() can be called
495 from interrupt context: either a hardware inte 495 from interrupt context: either a hardware interrupt or a softirq. An
496 example would be a timer which deletes object 496 example would be a timer which deletes object from the cache.
497 497
498 The change is shown below, in standard patch f 498 The change is shown below, in standard patch format: the ``-`` are lines
499 which are taken away, and the ``+`` are lines 499 which are taken away, and the ``+`` are lines which are added.
500 500
501 :: 501 ::
502 502
503 --- cache.c.usercontext 2003-12-09 13:58:5 503 --- cache.c.usercontext 2003-12-09 13:58:54.000000000 +1100
504 +++ cache.c.interrupt 2003-12-09 14:07:4 504 +++ cache.c.interrupt 2003-12-09 14:07:49.000000000 +1100
505 @@ -12,7 +12,7 @@ 505 @@ -12,7 +12,7 @@
506 int popularity; 506 int popularity;
507 }; 507 };
508 508
509 -static DEFINE_MUTEX(cache_lock); 509 -static DEFINE_MUTEX(cache_lock);
510 +static DEFINE_SPINLOCK(cache_lock); 510 +static DEFINE_SPINLOCK(cache_lock);
511 static LIST_HEAD(cache); 511 static LIST_HEAD(cache);
512 static unsigned int cache_num = 0; 512 static unsigned int cache_num = 0;
513 #define MAX_CACHE_SIZE 10 513 #define MAX_CACHE_SIZE 10
514 @@ -55,6 +55,7 @@ 514 @@ -55,6 +55,7 @@
515 int cache_add(int id, const char *name) 515 int cache_add(int id, const char *name)
516 { 516 {
517 struct object *obj; 517 struct object *obj;
518 + unsigned long flags; 518 + unsigned long flags;
519 519
520 if ((obj = kmalloc(sizeof(*obj), 520 if ((obj = kmalloc(sizeof(*obj), GFP_KERNEL)) == NULL)
521 return -ENOMEM; 521 return -ENOMEM;
522 @@ -63,30 +64,33 @@ 522 @@ -63,30 +64,33 @@
523 obj->id = id; 523 obj->id = id;
524 obj->popularity = 0; 524 obj->popularity = 0;
525 525
526 - mutex_lock(&cache_lock); 526 - mutex_lock(&cache_lock);
527 + spin_lock_irqsave(&cache_lock, fl 527 + spin_lock_irqsave(&cache_lock, flags);
528 __cache_add(obj); 528 __cache_add(obj);
529 - mutex_unlock(&cache_lock); 529 - mutex_unlock(&cache_lock);
530 + spin_unlock_irqrestore(&cache_loc 530 + spin_unlock_irqrestore(&cache_lock, flags);
531 return 0; 531 return 0;
532 } 532 }
533 533
534 void cache_delete(int id) 534 void cache_delete(int id)
535 { 535 {
536 - mutex_lock(&cache_lock); 536 - mutex_lock(&cache_lock);
537 + unsigned long flags; 537 + unsigned long flags;
538 + 538 +
539 + spin_lock_irqsave(&cache_lock, fl 539 + spin_lock_irqsave(&cache_lock, flags);
540 __cache_delete(__cache_find(id)); 540 __cache_delete(__cache_find(id));
541 - mutex_unlock(&cache_lock); 541 - mutex_unlock(&cache_lock);
542 + spin_unlock_irqrestore(&cache_loc 542 + spin_unlock_irqrestore(&cache_lock, flags);
543 } 543 }
544 544
545 int cache_find(int id, char *name) 545 int cache_find(int id, char *name)
546 { 546 {
547 struct object *obj; 547 struct object *obj;
548 int ret = -ENOENT; 548 int ret = -ENOENT;
549 + unsigned long flags; 549 + unsigned long flags;
550 550
551 - mutex_lock(&cache_lock); 551 - mutex_lock(&cache_lock);
552 + spin_lock_irqsave(&cache_lock, fl 552 + spin_lock_irqsave(&cache_lock, flags);
553 obj = __cache_find(id); 553 obj = __cache_find(id);
554 if (obj) { 554 if (obj) {
555 ret = 0; 555 ret = 0;
556 strcpy(name, obj->name); 556 strcpy(name, obj->name);
557 } 557 }
558 - mutex_unlock(&cache_lock); 558 - mutex_unlock(&cache_lock);
559 + spin_unlock_irqrestore(&cache_loc 559 + spin_unlock_irqrestore(&cache_lock, flags);
560 return ret; 560 return ret;
561 } 561 }
562 562
563 Note that the spin_lock_irqsave() will turn of 563 Note that the spin_lock_irqsave() will turn off
564 interrupts if they are on, otherwise does noth 564 interrupts if they are on, otherwise does nothing (if we are already in
565 an interrupt handler), hence these functions a 565 an interrupt handler), hence these functions are safe to call from any
566 context. 566 context.
567 567
568 Unfortunately, cache_add() calls kmalloc() 568 Unfortunately, cache_add() calls kmalloc()
569 with the ``GFP_KERNEL`` flag, which is only le 569 with the ``GFP_KERNEL`` flag, which is only legal in user context. I
570 have assumed that cache_add() is still only ca 570 have assumed that cache_add() is still only called in
571 user context, otherwise this should become a p 571 user context, otherwise this should become a parameter to
572 cache_add(). 572 cache_add().
573 573
574 Exposing Objects Outside This File 574 Exposing Objects Outside This File
575 ---------------------------------- 575 ----------------------------------
576 576
577 If our objects contained more information, it 577 If our objects contained more information, it might not be sufficient to
578 copy the information in and out: other parts o 578 copy the information in and out: other parts of the code might want to
579 keep pointers to these objects, for example, r 579 keep pointers to these objects, for example, rather than looking up the
580 id every time. This produces two problems. 580 id every time. This produces two problems.
581 581
582 The first problem is that we use the ``cache_l 582 The first problem is that we use the ``cache_lock`` to protect objects:
583 we'd need to make this non-static so the rest 583 we'd need to make this non-static so the rest of the code can use it.
584 This makes locking trickier, as it is no longe 584 This makes locking trickier, as it is no longer all in one place.
585 585
586 The second problem is the lifetime problem: if 586 The second problem is the lifetime problem: if another structure keeps a
587 pointer to an object, it presumably expects th 587 pointer to an object, it presumably expects that pointer to remain
588 valid. Unfortunately, this is only guaranteed 588 valid. Unfortunately, this is only guaranteed while you hold the lock,
589 otherwise someone might call cache_delete() an 589 otherwise someone might call cache_delete() and even
590 worse, add another object, re-using the same a 590 worse, add another object, re-using the same address.
591 591
592 As there is only one lock, you can't hold it f 592 As there is only one lock, you can't hold it forever: no-one else would
593 get any work done. 593 get any work done.
594 594
595 The solution to this problem is to use a refer 595 The solution to this problem is to use a reference count: everyone who
596 has a pointer to the object increases it when 596 has a pointer to the object increases it when they first get the object,
597 and drops the reference count when they're fin 597 and drops the reference count when they're finished with it. Whoever
598 drops it to zero knows it is unused, and can a 598 drops it to zero knows it is unused, and can actually delete it.
599 599
600 Here is the code:: 600 Here is the code::
601 601
602 --- cache.c.interrupt 2003-12-09 14:25:4 602 --- cache.c.interrupt 2003-12-09 14:25:43.000000000 +1100
603 +++ cache.c.refcnt 2003-12-09 14:33:05.00 603 +++ cache.c.refcnt 2003-12-09 14:33:05.000000000 +1100
604 @@ -7,6 +7,7 @@ 604 @@ -7,6 +7,7 @@
605 struct object 605 struct object
606 { 606 {
607 struct list_head list; 607 struct list_head list;
608 + unsigned int refcnt; 608 + unsigned int refcnt;
609 int id; 609 int id;
610 char name[32]; 610 char name[32];
611 int popularity; 611 int popularity;
612 @@ -17,6 +18,35 @@ 612 @@ -17,6 +18,35 @@
613 static unsigned int cache_num = 0; 613 static unsigned int cache_num = 0;
614 #define MAX_CACHE_SIZE 10 614 #define MAX_CACHE_SIZE 10
615 615
616 +static void __object_put(struct object *o 616 +static void __object_put(struct object *obj)
617 +{ 617 +{
618 + if (--obj->refcnt == 0) 618 + if (--obj->refcnt == 0)
619 + kfree(obj); 619 + kfree(obj);
620 +} 620 +}
621 + 621 +
622 +static void __object_get(struct object *o 622 +static void __object_get(struct object *obj)
623 +{ 623 +{
624 + obj->refcnt++; 624 + obj->refcnt++;
625 +} 625 +}
626 + 626 +
627 +void object_put(struct object *obj) 627 +void object_put(struct object *obj)
628 +{ 628 +{
629 + unsigned long flags; 629 + unsigned long flags;
630 + 630 +
631 + spin_lock_irqsave(&cache_lock, fl 631 + spin_lock_irqsave(&cache_lock, flags);
632 + __object_put(obj); 632 + __object_put(obj);
633 + spin_unlock_irqrestore(&cache_loc 633 + spin_unlock_irqrestore(&cache_lock, flags);
634 +} 634 +}
635 + 635 +
636 +void object_get(struct object *obj) 636 +void object_get(struct object *obj)
637 +{ 637 +{
638 + unsigned long flags; 638 + unsigned long flags;
639 + 639 +
640 + spin_lock_irqsave(&cache_lock, fl 640 + spin_lock_irqsave(&cache_lock, flags);
641 + __object_get(obj); 641 + __object_get(obj);
642 + spin_unlock_irqrestore(&cache_loc 642 + spin_unlock_irqrestore(&cache_lock, flags);
643 +} 643 +}
644 + 644 +
645 /* Must be holding cache_lock */ 645 /* Must be holding cache_lock */
646 static struct object *__cache_find(int id 646 static struct object *__cache_find(int id)
647 { 647 {
648 @@ -35,6 +65,7 @@ 648 @@ -35,6 +65,7 @@
649 { 649 {
650 BUG_ON(!obj); 650 BUG_ON(!obj);
651 list_del(&obj->list); 651 list_del(&obj->list);
652 + __object_put(obj); 652 + __object_put(obj);
653 cache_num--; 653 cache_num--;
654 } 654 }
655 655
656 @@ -63,6 +94,7 @@ 656 @@ -63,6 +94,7 @@
657 strscpy(obj->name, name, sizeof(o 657 strscpy(obj->name, name, sizeof(obj->name));
658 obj->id = id; 658 obj->id = id;
659 obj->popularity = 0; 659 obj->popularity = 0;
660 + obj->refcnt = 1; /* The cache hol 660 + obj->refcnt = 1; /* The cache holds a reference */
661 661
662 spin_lock_irqsave(&cache_lock, fl 662 spin_lock_irqsave(&cache_lock, flags);
663 __cache_add(obj); 663 __cache_add(obj);
664 @@ -79,18 +111,15 @@ 664 @@ -79,18 +111,15 @@
665 spin_unlock_irqrestore(&cache_loc 665 spin_unlock_irqrestore(&cache_lock, flags);
666 } 666 }
667 667
668 -int cache_find(int id, char *name) 668 -int cache_find(int id, char *name)
669 +struct object *cache_find(int id) 669 +struct object *cache_find(int id)
670 { 670 {
671 struct object *obj; 671 struct object *obj;
672 - int ret = -ENOENT; 672 - int ret = -ENOENT;
673 unsigned long flags; 673 unsigned long flags;
674 674
675 spin_lock_irqsave(&cache_lock, fl 675 spin_lock_irqsave(&cache_lock, flags);
676 obj = __cache_find(id); 676 obj = __cache_find(id);
677 - if (obj) { 677 - if (obj) {
678 - ret = 0; 678 - ret = 0;
679 - strcpy(name, obj->name); 679 - strcpy(name, obj->name);
680 - } 680 - }
681 + if (obj) 681 + if (obj)
682 + __object_get(obj); 682 + __object_get(obj);
683 spin_unlock_irqrestore(&cache_loc 683 spin_unlock_irqrestore(&cache_lock, flags);
684 - return ret; 684 - return ret;
685 + return obj; 685 + return obj;
686 } 686 }
687 687
688 We encapsulate the reference counting in the s 688 We encapsulate the reference counting in the standard 'get' and 'put'
689 functions. Now we can return the object itself 689 functions. Now we can return the object itself from
690 cache_find() which has the advantage that the 690 cache_find() which has the advantage that the user can
691 now sleep holding the object (eg. to copy_to_u 691 now sleep holding the object (eg. to copy_to_user() to
692 name to userspace). 692 name to userspace).
693 693
694 The other point to note is that I said a refer 694 The other point to note is that I said a reference should be held for
695 every pointer to the object: thus the referenc 695 every pointer to the object: thus the reference count is 1 when first
696 inserted into the cache. In some versions the 696 inserted into the cache. In some versions the framework does not hold a
697 reference count, but they are more complicated 697 reference count, but they are more complicated.
698 698
699 Using Atomic Operations For The Reference Coun 699 Using Atomic Operations For The Reference Count
700 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 700 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
701 701
702 In practice, :c:type:`atomic_t` would usually 702 In practice, :c:type:`atomic_t` would usually be used for refcnt. There are a
703 number of atomic operations defined in ``inclu 703 number of atomic operations defined in ``include/asm/atomic.h``: these
704 are guaranteed to be seen atomically from all 704 are guaranteed to be seen atomically from all CPUs in the system, so no
705 lock is required. In this case, it is simpler 705 lock is required. In this case, it is simpler than using spinlocks,
706 although for anything non-trivial using spinlo 706 although for anything non-trivial using spinlocks is clearer. The
707 atomic_inc() and atomic_dec_and_test() 707 atomic_inc() and atomic_dec_and_test()
708 are used instead of the standard increment and 708 are used instead of the standard increment and decrement operators, and
709 the lock is no longer used to protect the refe 709 the lock is no longer used to protect the reference count itself.
710 710
711 :: 711 ::
712 712
713 --- cache.c.refcnt 2003-12-09 15:00:35.00 713 --- cache.c.refcnt 2003-12-09 15:00:35.000000000 +1100
714 +++ cache.c.refcnt-atomic 2003-12-11 15: 714 +++ cache.c.refcnt-atomic 2003-12-11 15:49:42.000000000 +1100
715 @@ -7,7 +7,7 @@ 715 @@ -7,7 +7,7 @@
716 struct object 716 struct object
717 { 717 {
718 struct list_head list; 718 struct list_head list;
719 - unsigned int refcnt; 719 - unsigned int refcnt;
720 + atomic_t refcnt; 720 + atomic_t refcnt;
721 int id; 721 int id;
722 char name[32]; 722 char name[32];
723 int popularity; 723 int popularity;
724 @@ -18,33 +18,15 @@ 724 @@ -18,33 +18,15 @@
725 static unsigned int cache_num = 0; 725 static unsigned int cache_num = 0;
726 #define MAX_CACHE_SIZE 10 726 #define MAX_CACHE_SIZE 10
727 727
728 -static void __object_put(struct object *o 728 -static void __object_put(struct object *obj)
729 -{ 729 -{
730 - if (--obj->refcnt == 0) 730 - if (--obj->refcnt == 0)
731 - kfree(obj); 731 - kfree(obj);
732 -} 732 -}
733 - 733 -
734 -static void __object_get(struct object *o 734 -static void __object_get(struct object *obj)
735 -{ 735 -{
736 - obj->refcnt++; 736 - obj->refcnt++;
737 -} 737 -}
738 - 738 -
739 void object_put(struct object *obj) 739 void object_put(struct object *obj)
740 { 740 {
741 - unsigned long flags; 741 - unsigned long flags;
742 - 742 -
743 - spin_lock_irqsave(&cache_lock, fl 743 - spin_lock_irqsave(&cache_lock, flags);
744 - __object_put(obj); 744 - __object_put(obj);
745 - spin_unlock_irqrestore(&cache_loc 745 - spin_unlock_irqrestore(&cache_lock, flags);
746 + if (atomic_dec_and_test(&obj->ref 746 + if (atomic_dec_and_test(&obj->refcnt))
747 + kfree(obj); 747 + kfree(obj);
748 } 748 }
749 749
750 void object_get(struct object *obj) 750 void object_get(struct object *obj)
751 { 751 {
752 - unsigned long flags; 752 - unsigned long flags;
753 - 753 -
754 - spin_lock_irqsave(&cache_lock, fl 754 - spin_lock_irqsave(&cache_lock, flags);
755 - __object_get(obj); 755 - __object_get(obj);
756 - spin_unlock_irqrestore(&cache_loc 756 - spin_unlock_irqrestore(&cache_lock, flags);
757 + atomic_inc(&obj->refcnt); 757 + atomic_inc(&obj->refcnt);
758 } 758 }
759 759
760 /* Must be holding cache_lock */ 760 /* Must be holding cache_lock */
761 @@ -65,7 +47,7 @@ 761 @@ -65,7 +47,7 @@
762 { 762 {
763 BUG_ON(!obj); 763 BUG_ON(!obj);
764 list_del(&obj->list); 764 list_del(&obj->list);
765 - __object_put(obj); 765 - __object_put(obj);
766 + object_put(obj); 766 + object_put(obj);
767 cache_num--; 767 cache_num--;
768 } 768 }
769 769
770 @@ -94,7 +76,7 @@ 770 @@ -94,7 +76,7 @@
771 strscpy(obj->name, name, sizeof(o 771 strscpy(obj->name, name, sizeof(obj->name));
772 obj->id = id; 772 obj->id = id;
773 obj->popularity = 0; 773 obj->popularity = 0;
774 - obj->refcnt = 1; /* The cache hol 774 - obj->refcnt = 1; /* The cache holds a reference */
775 + atomic_set(&obj->refcnt, 1); /* T 775 + atomic_set(&obj->refcnt, 1); /* The cache holds a reference */
776 776
777 spin_lock_irqsave(&cache_lock, fl 777 spin_lock_irqsave(&cache_lock, flags);
778 __cache_add(obj); 778 __cache_add(obj);
779 @@ -119,7 +101,7 @@ 779 @@ -119,7 +101,7 @@
780 spin_lock_irqsave(&cache_lock, fl 780 spin_lock_irqsave(&cache_lock, flags);
781 obj = __cache_find(id); 781 obj = __cache_find(id);
782 if (obj) 782 if (obj)
783 - __object_get(obj); 783 - __object_get(obj);
784 + object_get(obj); 784 + object_get(obj);
785 spin_unlock_irqrestore(&cache_loc 785 spin_unlock_irqrestore(&cache_lock, flags);
786 return obj; 786 return obj;
787 } 787 }
788 788
789 Protecting The Objects Themselves 789 Protecting The Objects Themselves
790 --------------------------------- 790 ---------------------------------
791 791
792 In these examples, we assumed that the objects 792 In these examples, we assumed that the objects (except the reference
793 counts) never changed once they are created. I 793 counts) never changed once they are created. If we wanted to allow the
794 name to change, there are three possibilities: 794 name to change, there are three possibilities:
795 795
796 - You can make ``cache_lock`` non-static, and 796 - You can make ``cache_lock`` non-static, and tell people to grab that
797 lock before changing the name in any object 797 lock before changing the name in any object.
798 798
799 - You can provide a cache_obj_rename() which 799 - You can provide a cache_obj_rename() which grabs this
800 lock and changes the name for the caller, a 800 lock and changes the name for the caller, and tell everyone to use
801 that function. 801 that function.
802 802
803 - You can make the ``cache_lock`` protect onl 803 - You can make the ``cache_lock`` protect only the cache itself, and
804 use another lock to protect the name. 804 use another lock to protect the name.
805 805
806 Theoretically, you can make the locks as fine- 806 Theoretically, you can make the locks as fine-grained as one lock for
807 every field, for every object. In practice, th 807 every field, for every object. In practice, the most common variants
808 are: 808 are:
809 809
810 - One lock which protects the infrastructure 810 - One lock which protects the infrastructure (the ``cache`` list in
811 this example) and all the objects. This is 811 this example) and all the objects. This is what we have done so far.
812 812
813 - One lock which protects the infrastructure 813 - One lock which protects the infrastructure (including the list
814 pointers inside the objects), and one lock 814 pointers inside the objects), and one lock inside the object which
815 protects the rest of that object. 815 protects the rest of that object.
816 816
817 - Multiple locks to protect the infrastructur 817 - Multiple locks to protect the infrastructure (eg. one lock per hash
818 chain), possibly with a separate per-object 818 chain), possibly with a separate per-object lock.
819 819
820 Here is the "lock-per-object" implementation: 820 Here is the "lock-per-object" implementation:
821 821
822 :: 822 ::
823 823
824 --- cache.c.refcnt-atomic 2003-12-11 15: 824 --- cache.c.refcnt-atomic 2003-12-11 15:50:54.000000000 +1100
825 +++ cache.c.perobjectlock 2003-12-11 17: 825 +++ cache.c.perobjectlock 2003-12-11 17:15:03.000000000 +1100
826 @@ -6,11 +6,17 @@ 826 @@ -6,11 +6,17 @@
827 827
828 struct object 828 struct object
829 { 829 {
830 + /* These two protected by cache_l 830 + /* These two protected by cache_lock. */
831 struct list_head list; 831 struct list_head list;
832 + int popularity; 832 + int popularity;
833 + 833 +
834 atomic_t refcnt; 834 atomic_t refcnt;
835 + 835 +
836 + /* Doesn't change once created. * 836 + /* Doesn't change once created. */
837 int id; 837 int id;
838 + 838 +
839 + spinlock_t lock; /* Protects the 839 + spinlock_t lock; /* Protects the name */
840 char name[32]; 840 char name[32];
841 - int popularity; 841 - int popularity;
842 }; 842 };
843 843
844 static DEFINE_SPINLOCK(cache_lock); 844 static DEFINE_SPINLOCK(cache_lock);
845 @@ -77,6 +84,7 @@ 845 @@ -77,6 +84,7 @@
846 obj->id = id; 846 obj->id = id;
847 obj->popularity = 0; 847 obj->popularity = 0;
848 atomic_set(&obj->refcnt, 1); /* T 848 atomic_set(&obj->refcnt, 1); /* The cache holds a reference */
849 + spin_lock_init(&obj->lock); 849 + spin_lock_init(&obj->lock);
850 850
851 spin_lock_irqsave(&cache_lock, fl 851 spin_lock_irqsave(&cache_lock, flags);
852 __cache_add(obj); 852 __cache_add(obj);
853 853
854 Note that I decide that the popularity count s 854 Note that I decide that the popularity count should be protected by the
855 ``cache_lock`` rather than the per-object lock 855 ``cache_lock`` rather than the per-object lock: this is because it (like
856 the :c:type:`struct list_head <list_head>` ins 856 the :c:type:`struct list_head <list_head>` inside the object)
857 is logically part of the infrastructure. This 857 is logically part of the infrastructure. This way, I don't need to grab
858 the lock of every object in __cache_add() when 858 the lock of every object in __cache_add() when seeking
859 the least popular. 859 the least popular.
860 860
861 I also decided that the id member is unchangea 861 I also decided that the id member is unchangeable, so I don't need to
862 grab each object lock in __cache_find() to exa 862 grab each object lock in __cache_find() to examine the
863 id: the object lock is only used by a caller w 863 id: the object lock is only used by a caller who wants to read or write
864 the name field. 864 the name field.
865 865
866 Note also that I added a comment describing wh 866 Note also that I added a comment describing what data was protected by
867 which locks. This is extremely important, as i 867 which locks. This is extremely important, as it describes the runtime
868 behavior of the code, and can be hard to gain 868 behavior of the code, and can be hard to gain from just reading. And as
869 Alan Cox says, “Lock data, not code”. 869 Alan Cox says, “Lock data, not code”.
870 870
871 Common Problems 871 Common Problems
872 =============== 872 ===============
873 873
874 Deadlock: Simple and Advanced 874 Deadlock: Simple and Advanced
875 ----------------------------- 875 -----------------------------
876 876
877 There is a coding bug where a piece of code tr 877 There is a coding bug where a piece of code tries to grab a spinlock
878 twice: it will spin forever, waiting for the l 878 twice: it will spin forever, waiting for the lock to be released
879 (spinlocks, rwlocks and mutexes are not recurs 879 (spinlocks, rwlocks and mutexes are not recursive in Linux). This is
880 trivial to diagnose: not a 880 trivial to diagnose: not a
881 stay-up-five-nights-talk-to-fluffy-code-bunnie 881 stay-up-five-nights-talk-to-fluffy-code-bunnies kind of problem.
882 882
883 For a slightly more complex case, imagine you 883 For a slightly more complex case, imagine you have a region shared by a
884 softirq and user context. If you use a spin_lo 884 softirq and user context. If you use a spin_lock() call
885 to protect it, it is possible that the user co 885 to protect it, it is possible that the user context will be interrupted
886 by the softirq while it holds the lock, and th 886 by the softirq while it holds the lock, and the softirq will then spin
887 forever trying to get the same lock. 887 forever trying to get the same lock.
888 888
889 Both of these are called deadlock, and as show 889 Both of these are called deadlock, and as shown above, it can occur even
890 with a single CPU (although not on UP compiles 890 with a single CPU (although not on UP compiles, since spinlocks vanish
891 on kernel compiles with ``CONFIG_SMP``\ =n. Yo 891 on kernel compiles with ``CONFIG_SMP``\ =n. You'll still get data
892 corruption in the second example). 892 corruption in the second example).
893 893
894 This complete lockup is easy to diagnose: on S 894 This complete lockup is easy to diagnose: on SMP boxes the watchdog
895 timer or compiling with ``DEBUG_SPINLOCK`` set 895 timer or compiling with ``DEBUG_SPINLOCK`` set
896 (``include/linux/spinlock.h``) will show this 896 (``include/linux/spinlock.h``) will show this up immediately when it
897 happens. 897 happens.
898 898
899 A more complex problem is the so-called 'deadl 899 A more complex problem is the so-called 'deadly embrace', involving two
900 or more locks. Say you have a hash table: each 900 or more locks. Say you have a hash table: each entry in the table is a
901 spinlock, and a chain of hashed objects. Insid 901 spinlock, and a chain of hashed objects. Inside a softirq handler, you
902 sometimes want to alter an object from one pla 902 sometimes want to alter an object from one place in the hash to another:
903 you grab the spinlock of the old hash chain an 903 you grab the spinlock of the old hash chain and the spinlock of the new
904 hash chain, and delete the object from the old 904 hash chain, and delete the object from the old one, and insert it in the
905 new one. 905 new one.
906 906
907 There are two problems here. First, if your co 907 There are two problems here. First, if your code ever tries to move the
908 object to the same chain, it will deadlock wit 908 object to the same chain, it will deadlock with itself as it tries to
909 lock it twice. Secondly, if the same softirq o 909 lock it twice. Secondly, if the same softirq on another CPU is trying to
910 move another object in the reverse direction, 910 move another object in the reverse direction, the following could
911 happen: 911 happen:
912 912
913 +-----------------------+--------------------- 913 +-----------------------+-----------------------+
914 | CPU 1 | CPU 2 914 | CPU 1 | CPU 2 |
915 +=======================+===================== 915 +=======================+=======================+
916 | Grab lock A -> OK | Grab lock B -> OK 916 | Grab lock A -> OK | Grab lock B -> OK |
917 +-----------------------+--------------------- 917 +-----------------------+-----------------------+
918 | Grab lock B -> spin | Grab lock A -> spin 918 | Grab lock B -> spin | Grab lock A -> spin |
919 +-----------------------+--------------------- 919 +-----------------------+-----------------------+
920 920
921 Table: Consequences 921 Table: Consequences
922 922
923 The two CPUs will spin forever, waiting for th 923 The two CPUs will spin forever, waiting for the other to give up their
924 lock. It will look, smell, and feel like a cra 924 lock. It will look, smell, and feel like a crash.
925 925
926 Preventing Deadlock 926 Preventing Deadlock
927 ------------------- 927 -------------------
928 928
929 Textbooks will tell you that if you always loc 929 Textbooks will tell you that if you always lock in the same order, you
930 will never get this kind of deadlock. Practice 930 will never get this kind of deadlock. Practice will tell you that this
931 approach doesn't scale: when I create a new lo 931 approach doesn't scale: when I create a new lock, I don't understand
932 enough of the kernel to figure out where in th 932 enough of the kernel to figure out where in the 5000 lock hierarchy it
933 will fit. 933 will fit.
934 934
935 The best locks are encapsulated: they never ge 935 The best locks are encapsulated: they never get exposed in headers, and
936 are never held around calls to non-trivial fun 936 are never held around calls to non-trivial functions outside the same
937 file. You can read through this code and see t 937 file. You can read through this code and see that it will never
938 deadlock, because it never tries to grab anoth 938 deadlock, because it never tries to grab another lock while it has that
939 one. People using your code don't even need to 939 one. People using your code don't even need to know you are using a
940 lock. 940 lock.
941 941
942 A classic problem here is when you provide cal 942 A classic problem here is when you provide callbacks or hooks: if you
943 call these with the lock held, you risk simple 943 call these with the lock held, you risk simple deadlock, or a deadly
944 embrace (who knows what the callback will do?) !! 944 embrace (who knows what the callback will do?). Remember, the other
>> 945 programmers are out to get you, so don't do this.
945 946
946 Overzealous Prevention Of Deadlocks 947 Overzealous Prevention Of Deadlocks
947 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 948 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
948 949
949 Deadlocks are problematic, but not as bad as d 950 Deadlocks are problematic, but not as bad as data corruption. Code which
950 grabs a read lock, searches a list, fails to f 951 grabs a read lock, searches a list, fails to find what it wants, drops
951 the read lock, grabs a write lock and inserts 952 the read lock, grabs a write lock and inserts the object has a race
952 condition. 953 condition.
953 954
>> 955 If you don't see why, please stay away from my code.
>> 956
954 Racing Timers: A Kernel Pastime 957 Racing Timers: A Kernel Pastime
955 ------------------------------- 958 -------------------------------
956 959
957 Timers can produce their own special problems 960 Timers can produce their own special problems with races. Consider a
958 collection of objects (list, hash, etc) where 961 collection of objects (list, hash, etc) where each object has a timer
959 which is due to destroy it. 962 which is due to destroy it.
960 963
961 If you want to destroy the entire collection ( 964 If you want to destroy the entire collection (say on module removal),
962 you might do the following:: 965 you might do the following::
963 966
964 /* THIS CODE BAD BAD BAD BAD: IF I 967 /* THIS CODE BAD BAD BAD BAD: IF IT WAS ANY WORSE IT WOULD USE
965 HUNGARIAN NOTATION */ 968 HUNGARIAN NOTATION */
966 spin_lock_bh(&list_lock); 969 spin_lock_bh(&list_lock);
967 970
968 while (list) { 971 while (list) {
969 struct foo *next = list->n 972 struct foo *next = list->next;
970 timer_delete(&list->timer) !! 973 del_timer(&list->timer);
971 kfree(list); 974 kfree(list);
972 list = next; 975 list = next;
973 } 976 }
974 977
975 spin_unlock_bh(&list_lock); 978 spin_unlock_bh(&list_lock);
976 979
977 980
978 Sooner or later, this will crash on SMP, becau 981 Sooner or later, this will crash on SMP, because a timer can have just
979 gone off before the spin_lock_bh(), and it wil 982 gone off before the spin_lock_bh(), and it will only get
980 the lock after we spin_unlock_bh(), and then t 983 the lock after we spin_unlock_bh(), and then try to free
981 the element (which has already been freed!). 984 the element (which has already been freed!).
982 985
983 This can be avoided by checking the result of 986 This can be avoided by checking the result of
984 timer_delete(): if it returns 1, the timer has !! 987 del_timer(): if it returns 1, the timer has been deleted.
985 If 0, it means (in this case) that it is curre 988 If 0, it means (in this case) that it is currently running, so we can
986 do:: 989 do::
987 990
988 retry: 991 retry:
989 spin_lock_bh(&list_lock); 992 spin_lock_bh(&list_lock);
990 993
991 while (list) { 994 while (list) {
992 struct foo *next = 995 struct foo *next = list->next;
993 if (!timer_delete( !! 996 if (!del_timer(&list->timer)) {
994 /* Give ti 997 /* Give timer a chance to delete this */
995 spin_unloc 998 spin_unlock_bh(&list_lock);
996 goto retry 999 goto retry;
997 } 1000 }
998 kfree(list); 1001 kfree(list);
999 list = next; 1002 list = next;
1000 } 1003 }
1001 1004
1002 spin_unlock_bh(&list_lock 1005 spin_unlock_bh(&list_lock);
1003 1006
1004 1007
1005 Another common problem is deleting timers whi 1008 Another common problem is deleting timers which restart themselves (by
1006 calling add_timer() at the end of their timer 1009 calling add_timer() at the end of their timer function).
1007 Because this is a fairly common case which is 1010 Because this is a fairly common case which is prone to races, you should
1008 use timer_delete_sync() (``include/linux/time !! 1011 use del_timer_sync() (``include/linux/timer.h``) to
1009 !! 1012 handle this case. It returns the number of times the timer had to be
1010 Before freeing a timer, timer_shutdown() or t !! 1013 deleted before we finally stopped it from adding itself back in.
1011 called which will keep it from being rearmed. <<
1012 rearm the timer will be silently ignored by t <<
1013 <<
1014 1014
1015 Locking Speed 1015 Locking Speed
1016 ============= 1016 =============
1017 1017
1018 There are three main things to worry about wh 1018 There are three main things to worry about when considering speed of
1019 some code which does locking. First is concur 1019 some code which does locking. First is concurrency: how many things are
1020 going to be waiting while someone else is hol 1020 going to be waiting while someone else is holding a lock. Second is the
1021 time taken to actually acquire and release an 1021 time taken to actually acquire and release an uncontended lock. Third is
1022 using fewer, or smarter locks. I'm assuming t 1022 using fewer, or smarter locks. I'm assuming that the lock is used fairly
1023 often: otherwise, you wouldn't be concerned a 1023 often: otherwise, you wouldn't be concerned about efficiency.
1024 1024
1025 Concurrency depends on how long the lock is u 1025 Concurrency depends on how long the lock is usually held: you should
1026 hold the lock for as long as needed, but no l 1026 hold the lock for as long as needed, but no longer. In the cache
1027 example, we always create the object without 1027 example, we always create the object without the lock held, and then
1028 grab the lock only when we are ready to inser 1028 grab the lock only when we are ready to insert it in the list.
1029 1029
1030 Acquisition times depend on how much damage t 1030 Acquisition times depend on how much damage the lock operations do to
1031 the pipeline (pipeline stalls) and how likely 1031 the pipeline (pipeline stalls) and how likely it is that this CPU was
1032 the last one to grab the lock (ie. is the loc 1032 the last one to grab the lock (ie. is the lock cache-hot for this CPU):
1033 on a machine with more CPUs, this likelihood 1033 on a machine with more CPUs, this likelihood drops fast. Consider a
1034 700MHz Intel Pentium III: an instruction take 1034 700MHz Intel Pentium III: an instruction takes about 0.7ns, an atomic
1035 increment takes about 58ns, a lock which is c 1035 increment takes about 58ns, a lock which is cache-hot on this CPU takes
1036 160ns, and a cacheline transfer from another 1036 160ns, and a cacheline transfer from another CPU takes an additional 170
1037 to 360ns. (These figures from Paul McKenney's 1037 to 360ns. (These figures from Paul McKenney's `Linux Journal RCU
1038 article <http://www.linuxjournal.com/article. 1038 article <http://www.linuxjournal.com/article.php?sid=6993>`__).
1039 1039
1040 These two aims conflict: holding a lock for a 1040 These two aims conflict: holding a lock for a short time might be done
1041 by splitting locks into parts (such as in our 1041 by splitting locks into parts (such as in our final per-object-lock
1042 example), but this increases the number of lo 1042 example), but this increases the number of lock acquisitions, and the
1043 results are often slower than having a single 1043 results are often slower than having a single lock. This is another
1044 reason to advocate locking simplicity. 1044 reason to advocate locking simplicity.
1045 1045
1046 The third concern is addressed below: there a 1046 The third concern is addressed below: there are some methods to reduce
1047 the amount of locking which needs to be done. 1047 the amount of locking which needs to be done.
1048 1048
1049 Read/Write Lock Variants 1049 Read/Write Lock Variants
1050 ------------------------ 1050 ------------------------
1051 1051
1052 Both spinlocks and mutexes have read/write va 1052 Both spinlocks and mutexes have read/write variants: ``rwlock_t`` and
1053 :c:type:`struct rw_semaphore <rw_semaphore>`. 1053 :c:type:`struct rw_semaphore <rw_semaphore>`. These divide
1054 users into two classes: the readers and the w 1054 users into two classes: the readers and the writers. If you are only
1055 reading the data, you can get a read lock, bu 1055 reading the data, you can get a read lock, but to write to the data you
1056 need the write lock. Many people can hold a r 1056 need the write lock. Many people can hold a read lock, but a writer must
1057 be sole holder. 1057 be sole holder.
1058 1058
1059 If your code divides neatly along reader/writ 1059 If your code divides neatly along reader/writer lines (as our cache code
1060 does), and the lock is held by readers for si 1060 does), and the lock is held by readers for significant lengths of time,
1061 using these locks can help. They are slightly 1061 using these locks can help. They are slightly slower than the normal
1062 locks though, so in practice ``rwlock_t`` is 1062 locks though, so in practice ``rwlock_t`` is not usually worthwhile.
1063 1063
1064 Avoiding Locks: Read Copy Update 1064 Avoiding Locks: Read Copy Update
1065 -------------------------------- 1065 --------------------------------
1066 1066
1067 There is a special method of read/write locki 1067 There is a special method of read/write locking called Read Copy Update.
1068 Using RCU, the readers can avoid taking a loc 1068 Using RCU, the readers can avoid taking a lock altogether: as we expect
1069 our cache to be read more often than updated 1069 our cache to be read more often than updated (otherwise the cache is a
1070 waste of time), it is a candidate for this op 1070 waste of time), it is a candidate for this optimization.
1071 1071
1072 How do we get rid of read locks? Getting rid 1072 How do we get rid of read locks? Getting rid of read locks means that
1073 writers may be changing the list underneath t 1073 writers may be changing the list underneath the readers. That is
1074 actually quite simple: we can read a linked l 1074 actually quite simple: we can read a linked list while an element is
1075 being added if the writer adds the element ve 1075 being added if the writer adds the element very carefully. For example,
1076 adding ``new`` to a single linked list called 1076 adding ``new`` to a single linked list called ``list``::
1077 1077
1078 new->next = list->next; 1078 new->next = list->next;
1079 wmb(); 1079 wmb();
1080 list->next = new; 1080 list->next = new;
1081 1081
1082 1082
1083 The wmb() is a write memory barrier. It ensur 1083 The wmb() is a write memory barrier. It ensures that the
1084 first operation (setting the new element's `` 1084 first operation (setting the new element's ``next`` pointer) is complete
1085 and will be seen by all CPUs, before the seco 1085 and will be seen by all CPUs, before the second operation is (putting
1086 the new element into the list). This is impor 1086 the new element into the list). This is important, since modern
1087 compilers and modern CPUs can both reorder in 1087 compilers and modern CPUs can both reorder instructions unless told
1088 otherwise: we want a reader to either not see 1088 otherwise: we want a reader to either not see the new element at all, or
1089 see the new element with the ``next`` pointer 1089 see the new element with the ``next`` pointer correctly pointing at the
1090 rest of the list. 1090 rest of the list.
1091 1091
1092 Fortunately, there is a function to do this f 1092 Fortunately, there is a function to do this for standard
1093 :c:type:`struct list_head <list_head>` lists: 1093 :c:type:`struct list_head <list_head>` lists:
1094 list_add_rcu() (``include/linux/list.h``). 1094 list_add_rcu() (``include/linux/list.h``).
1095 1095
1096 Removing an element from the list is even sim 1096 Removing an element from the list is even simpler: we replace the
1097 pointer to the old element with a pointer to 1097 pointer to the old element with a pointer to its successor, and readers
1098 will either see it, or skip over it. 1098 will either see it, or skip over it.
1099 1099
1100 :: 1100 ::
1101 1101
1102 list->next = old->next; 1102 list->next = old->next;
1103 1103
1104 1104
1105 There is list_del_rcu() (``include/linux/list 1105 There is list_del_rcu() (``include/linux/list.h``) which
1106 does this (the normal version poisons the old 1106 does this (the normal version poisons the old object, which we don't
1107 want). 1107 want).
1108 1108
1109 The reader must also be careful: some CPUs ca 1109 The reader must also be careful: some CPUs can look through the ``next``
1110 pointer to start reading the contents of the 1110 pointer to start reading the contents of the next element early, but
1111 don't realize that the pre-fetched contents i 1111 don't realize that the pre-fetched contents is wrong when the ``next``
1112 pointer changes underneath them. Once again, 1112 pointer changes underneath them. Once again, there is a
1113 list_for_each_entry_rcu() (``include/linux/li 1113 list_for_each_entry_rcu() (``include/linux/list.h``)
1114 to help you. Of course, writers can just use 1114 to help you. Of course, writers can just use
1115 list_for_each_entry(), since there cannot be 1115 list_for_each_entry(), since there cannot be two
1116 simultaneous writers. 1116 simultaneous writers.
1117 1117
1118 Our final dilemma is this: when can we actual 1118 Our final dilemma is this: when can we actually destroy the removed
1119 element? Remember, a reader might be stepping 1119 element? Remember, a reader might be stepping through this element in
1120 the list right now: if we free this element a 1120 the list right now: if we free this element and the ``next`` pointer
1121 changes, the reader will jump off into garbag 1121 changes, the reader will jump off into garbage and crash. We need to
1122 wait until we know that all the readers who w 1122 wait until we know that all the readers who were traversing the list
1123 when we deleted the element are finished. We 1123 when we deleted the element are finished. We use
1124 call_rcu() to register a callback which will 1124 call_rcu() to register a callback which will actually
1125 destroy the object once all pre-existing read 1125 destroy the object once all pre-existing readers are finished.
1126 Alternatively, synchronize_rcu() may be used 1126 Alternatively, synchronize_rcu() may be used to block
1127 until all pre-existing are finished. 1127 until all pre-existing are finished.
1128 1128
1129 But how does Read Copy Update know when the r 1129 But how does Read Copy Update know when the readers are finished? The
1130 method is this: firstly, the readers always t 1130 method is this: firstly, the readers always traverse the list inside
1131 rcu_read_lock()/rcu_read_unlock() pairs: 1131 rcu_read_lock()/rcu_read_unlock() pairs:
1132 these simply disable preemption so the reader 1132 these simply disable preemption so the reader won't go to sleep while
1133 reading the list. 1133 reading the list.
1134 1134
1135 RCU then waits until every other CPU has slep 1135 RCU then waits until every other CPU has slept at least once: since
1136 readers cannot sleep, we know that any reader 1136 readers cannot sleep, we know that any readers which were traversing the
1137 list during the deletion are finished, and th 1137 list during the deletion are finished, and the callback is triggered.
1138 The real Read Copy Update code is a little mo 1138 The real Read Copy Update code is a little more optimized than this, but
1139 this is the fundamental idea. 1139 this is the fundamental idea.
1140 1140
1141 :: 1141 ::
1142 1142
1143 --- cache.c.perobjectlock 2003-12-11 17 1143 --- cache.c.perobjectlock 2003-12-11 17:15:03.000000000 +1100
1144 +++ cache.c.rcupdate 2003-12-11 17:55: 1144 +++ cache.c.rcupdate 2003-12-11 17:55:14.000000000 +1100
1145 @@ -1,15 +1,18 @@ 1145 @@ -1,15 +1,18 @@
1146 #include <linux/list.h> 1146 #include <linux/list.h>
1147 #include <linux/slab.h> 1147 #include <linux/slab.h>
1148 #include <linux/string.h> 1148 #include <linux/string.h>
1149 +#include <linux/rcupdate.h> 1149 +#include <linux/rcupdate.h>
1150 #include <linux/mutex.h> 1150 #include <linux/mutex.h>
1151 #include <asm/errno.h> 1151 #include <asm/errno.h>
1152 1152
1153 struct object 1153 struct object
1154 { 1154 {
1155 - /* These two protected by cache_ 1155 - /* These two protected by cache_lock. */
1156 + /* This is protected by RCU */ 1156 + /* This is protected by RCU */
1157 struct list_head list; 1157 struct list_head list;
1158 int popularity; 1158 int popularity;
1159 1159
1160 + struct rcu_head rcu; 1160 + struct rcu_head rcu;
1161 + 1161 +
1162 atomic_t refcnt; 1162 atomic_t refcnt;
1163 1163
1164 /* Doesn't change once created. 1164 /* Doesn't change once created. */
1165 @@ -40,7 +43,7 @@ 1165 @@ -40,7 +43,7 @@
1166 { 1166 {
1167 struct object *i; 1167 struct object *i;
1168 1168
1169 - list_for_each_entry(i, &cache, l 1169 - list_for_each_entry(i, &cache, list) {
1170 + list_for_each_entry_rcu(i, &cach 1170 + list_for_each_entry_rcu(i, &cache, list) {
1171 if (i->id == id) { 1171 if (i->id == id) {
1172 i->popularity++; 1172 i->popularity++;
1173 return i; 1173 return i;
1174 @@ -49,19 +52,25 @@ 1174 @@ -49,19 +52,25 @@
1175 return NULL; 1175 return NULL;
1176 } 1176 }
1177 1177
1178 +/* Final discard done once we know no re 1178 +/* Final discard done once we know no readers are looking. */
1179 +static void cache_delete_rcu(void *arg) 1179 +static void cache_delete_rcu(void *arg)
1180 +{ 1180 +{
1181 + object_put(arg); 1181 + object_put(arg);
1182 +} 1182 +}
1183 + 1183 +
1184 /* Must be holding cache_lock */ 1184 /* Must be holding cache_lock */
1185 static void __cache_delete(struct object 1185 static void __cache_delete(struct object *obj)
1186 { 1186 {
1187 BUG_ON(!obj); 1187 BUG_ON(!obj);
1188 - list_del(&obj->list); 1188 - list_del(&obj->list);
1189 - object_put(obj); 1189 - object_put(obj);
1190 + list_del_rcu(&obj->list); 1190 + list_del_rcu(&obj->list);
1191 cache_num--; 1191 cache_num--;
1192 + call_rcu(&obj->rcu, cache_delete 1192 + call_rcu(&obj->rcu, cache_delete_rcu);
1193 } 1193 }
1194 1194
1195 /* Must be holding cache_lock */ 1195 /* Must be holding cache_lock */
1196 static void __cache_add(struct object *o 1196 static void __cache_add(struct object *obj)
1197 { 1197 {
1198 - list_add(&obj->list, &cache); 1198 - list_add(&obj->list, &cache);
1199 + list_add_rcu(&obj->list, &cache) 1199 + list_add_rcu(&obj->list, &cache);
1200 if (++cache_num > MAX_CACHE_SIZE 1200 if (++cache_num > MAX_CACHE_SIZE) {
1201 struct object *i, *outca 1201 struct object *i, *outcast = NULL;
1202 list_for_each_entry(i, & 1202 list_for_each_entry(i, &cache, list) {
1203 @@ -104,12 +114,11 @@ 1203 @@ -104,12 +114,11 @@
1204 struct object *cache_find(int id) 1204 struct object *cache_find(int id)
1205 { 1205 {
1206 struct object *obj; 1206 struct object *obj;
1207 - unsigned long flags; 1207 - unsigned long flags;
1208 1208
1209 - spin_lock_irqsave(&cache_lock, f 1209 - spin_lock_irqsave(&cache_lock, flags);
1210 + rcu_read_lock(); 1210 + rcu_read_lock();
1211 obj = __cache_find(id); 1211 obj = __cache_find(id);
1212 if (obj) 1212 if (obj)
1213 object_get(obj); 1213 object_get(obj);
1214 - spin_unlock_irqrestore(&cache_lo 1214 - spin_unlock_irqrestore(&cache_lock, flags);
1215 + rcu_read_unlock(); 1215 + rcu_read_unlock();
1216 return obj; 1216 return obj;
1217 } 1217 }
1218 1218
1219 Note that the reader will alter the popularit 1219 Note that the reader will alter the popularity member in
1220 __cache_find(), and now it doesn't hold a loc 1220 __cache_find(), and now it doesn't hold a lock. One
1221 solution would be to make it an ``atomic_t``, 1221 solution would be to make it an ``atomic_t``, but for this usage, we
1222 don't really care about races: an approximate 1222 don't really care about races: an approximate result is good enough, so
1223 I didn't change it. 1223 I didn't change it.
1224 1224
1225 The result is that cache_find() requires no 1225 The result is that cache_find() requires no
1226 synchronization with any other functions, so 1226 synchronization with any other functions, so is almost as fast on SMP as
1227 it would be on UP. 1227 it would be on UP.
1228 1228
1229 There is a further optimization possible here 1229 There is a further optimization possible here: remember our original
1230 cache code, where there were no reference cou 1230 cache code, where there were no reference counts and the caller simply
1231 held the lock whenever using the object? This 1231 held the lock whenever using the object? This is still possible: if you
1232 hold the lock, no one can delete the object, 1232 hold the lock, no one can delete the object, so you don't need to get
1233 and put the reference count. 1233 and put the reference count.
1234 1234
1235 Now, because the 'read lock' in RCU is simply 1235 Now, because the 'read lock' in RCU is simply disabling preemption, a
1236 caller which always has preemption disabled b 1236 caller which always has preemption disabled between calling
1237 cache_find() and object_put() does not 1237 cache_find() and object_put() does not
1238 need to actually get and put the reference co 1238 need to actually get and put the reference count: we could expose
1239 __cache_find() by making it non-static, and s 1239 __cache_find() by making it non-static, and such
1240 callers could simply call that. 1240 callers could simply call that.
1241 1241
1242 The benefit here is that the reference count 1242 The benefit here is that the reference count is not written to: the
1243 object is not altered in any way, which is mu 1243 object is not altered in any way, which is much faster on SMP machines
1244 due to caching. 1244 due to caching.
1245 1245
1246 Per-CPU Data 1246 Per-CPU Data
1247 ------------ 1247 ------------
1248 1248
1249 Another technique for avoiding locking which 1249 Another technique for avoiding locking which is used fairly widely is to
1250 duplicate information for each CPU. For examp 1250 duplicate information for each CPU. For example, if you wanted to keep a
1251 count of a common condition, you could use a 1251 count of a common condition, you could use a spin lock and a single
1252 counter. Nice and simple. 1252 counter. Nice and simple.
1253 1253
1254 If that was too slow (it's usually not, but i 1254 If that was too slow (it's usually not, but if you've got a really big
1255 machine to test on and can show that it is), 1255 machine to test on and can show that it is), you could instead use a
1256 counter for each CPU, then none of them need 1256 counter for each CPU, then none of them need an exclusive lock. See
1257 DEFINE_PER_CPU(), get_cpu_var() and 1257 DEFINE_PER_CPU(), get_cpu_var() and
1258 put_cpu_var() (``include/linux/percpu.h``). 1258 put_cpu_var() (``include/linux/percpu.h``).
1259 1259
1260 Of particular use for simple per-cpu counters 1260 Of particular use for simple per-cpu counters is the ``local_t`` type,
1261 and the cpu_local_inc() and related functions 1261 and the cpu_local_inc() and related functions, which are
1262 more efficient than simple code on some archi 1262 more efficient than simple code on some architectures
1263 (``include/asm/local.h``). 1263 (``include/asm/local.h``).
1264 1264
1265 Note that there is no simple, reliable way of 1265 Note that there is no simple, reliable way of getting an exact value of
1266 such a counter, without introducing more lock 1266 such a counter, without introducing more locks. This is not a problem
1267 for some uses. 1267 for some uses.
1268 1268
1269 Data Which Mostly Used By An IRQ Handler 1269 Data Which Mostly Used By An IRQ Handler
1270 ---------------------------------------- 1270 ----------------------------------------
1271 1271
1272 If data is always accessed from within the sa 1272 If data is always accessed from within the same IRQ handler, you don't
1273 need a lock at all: the kernel already guaran 1273 need a lock at all: the kernel already guarantees that the irq handler
1274 will not run simultaneously on multiple CPUs. 1274 will not run simultaneously on multiple CPUs.
1275 1275
1276 Manfred Spraul points out that you can still 1276 Manfred Spraul points out that you can still do this, even if the data
1277 is very occasionally accessed in user context 1277 is very occasionally accessed in user context or softirqs/tasklets. The
1278 irq handler doesn't use a lock, and all other 1278 irq handler doesn't use a lock, and all other accesses are done as so::
1279 1279
1280 mutex_lock(&lock); !! 1280 spin_lock(&lock);
1281 disable_irq(irq); 1281 disable_irq(irq);
1282 ... 1282 ...
1283 enable_irq(irq); 1283 enable_irq(irq);
1284 mutex_unlock(&lock); !! 1284 spin_unlock(&lock);
1285 1285
1286 The disable_irq() prevents the irq handler fr 1286 The disable_irq() prevents the irq handler from running
1287 (and waits for it to finish if it's currently 1287 (and waits for it to finish if it's currently running on other CPUs).
1288 The spinlock prevents any other accesses happ 1288 The spinlock prevents any other accesses happening at the same time.
1289 Naturally, this is slower than just a spin_lo 1289 Naturally, this is slower than just a spin_lock_irq()
1290 call, so it only makes sense if this type of 1290 call, so it only makes sense if this type of access happens extremely
1291 rarely. 1291 rarely.
1292 1292
1293 What Functions Are Safe To Call From Interrup 1293 What Functions Are Safe To Call From Interrupts?
1294 ============================================= 1294 ================================================
1295 1295
1296 Many functions in the kernel sleep (ie. call 1296 Many functions in the kernel sleep (ie. call schedule()) directly or
1297 indirectly: you can never call them while hol 1297 indirectly: you can never call them while holding a spinlock, or with
1298 preemption disabled. This also means you need 1298 preemption disabled. This also means you need to be in user context:
1299 calling them from an interrupt is illegal. 1299 calling them from an interrupt is illegal.
1300 1300
1301 Some Functions Which Sleep 1301 Some Functions Which Sleep
1302 -------------------------- 1302 --------------------------
1303 1303
1304 The most common ones are listed below, but yo 1304 The most common ones are listed below, but you usually have to read the
1305 code to find out if other calls are safe. If 1305 code to find out if other calls are safe. If everyone else who calls it
1306 can sleep, you probably need to be able to sl 1306 can sleep, you probably need to be able to sleep, too. In particular,
1307 registration and deregistration functions usu 1307 registration and deregistration functions usually expect to be called
1308 from user context, and can sleep. 1308 from user context, and can sleep.
1309 1309
1310 - Accesses to userspace: 1310 - Accesses to userspace:
1311 1311
1312 - copy_from_user() 1312 - copy_from_user()
1313 1313
1314 - copy_to_user() 1314 - copy_to_user()
1315 1315
1316 - get_user() 1316 - get_user()
1317 1317
1318 - put_user() 1318 - put_user()
1319 1319
1320 - kmalloc(GP_KERNEL) <kmalloc>` 1320 - kmalloc(GP_KERNEL) <kmalloc>`
1321 1321
1322 - mutex_lock_interruptible() and 1322 - mutex_lock_interruptible() and
1323 mutex_lock() 1323 mutex_lock()
1324 1324
1325 There is a mutex_trylock() which does not 1325 There is a mutex_trylock() which does not sleep.
1326 Still, it must not be used inside interrup 1326 Still, it must not be used inside interrupt context since its
1327 implementation is not safe for that. mutex 1327 implementation is not safe for that. mutex_unlock()
1328 will also never sleep. It cannot be used i 1328 will also never sleep. It cannot be used in interrupt context either
1329 since a mutex must be released by the same 1329 since a mutex must be released by the same task that acquired it.
1330 1330
1331 Some Functions Which Don't Sleep 1331 Some Functions Which Don't Sleep
1332 -------------------------------- 1332 --------------------------------
1333 1333
1334 Some functions are safe to call from any cont 1334 Some functions are safe to call from any context, or holding almost any
1335 lock. 1335 lock.
1336 1336
1337 - printk() 1337 - printk()
1338 1338
1339 - kfree() 1339 - kfree()
1340 1340
1341 - add_timer() and timer_delete() !! 1341 - add_timer() and del_timer()
1342 1342
1343 Mutex API reference 1343 Mutex API reference
1344 =================== 1344 ===================
1345 1345
1346 .. kernel-doc:: include/linux/mutex.h 1346 .. kernel-doc:: include/linux/mutex.h
1347 :internal: 1347 :internal:
1348 1348
1349 .. kernel-doc:: kernel/locking/mutex.c 1349 .. kernel-doc:: kernel/locking/mutex.c
1350 :export: 1350 :export:
1351 1351
1352 Futex API reference 1352 Futex API reference
1353 =================== 1353 ===================
1354 1354
1355 .. kernel-doc:: kernel/futex/core.c 1355 .. kernel-doc:: kernel/futex/core.c
1356 :internal: <<
1357 <<
1358 .. kernel-doc:: kernel/futex/futex.h <<
1359 :internal: <<
1360 <<
1361 .. kernel-doc:: kernel/futex/pi.c <<
1362 :internal: <<
1363 <<
1364 .. kernel-doc:: kernel/futex/requeue.c <<
1365 :internal: <<
1366 <<
1367 .. kernel-doc:: kernel/futex/waitwake.c <<
1368 :internal: 1356 :internal:
1369 1357
1370 Further reading 1358 Further reading
1371 =============== 1359 ===============
1372 1360
1373 - ``Documentation/locking/spinlocks.rst``: L 1361 - ``Documentation/locking/spinlocks.rst``: Linus Torvalds' spinlocking
1374 tutorial in the kernel sources. 1362 tutorial in the kernel sources.
1375 1363
1376 - Unix Systems for Modern Architectures: Sym 1364 - Unix Systems for Modern Architectures: Symmetric Multiprocessing and
1377 Caching for Kernel Programmers: 1365 Caching for Kernel Programmers:
1378 1366
1379 Curt Schimmel's very good introduction to 1367 Curt Schimmel's very good introduction to kernel level locking (not
1380 written for Linux, but nearly everything a 1368 written for Linux, but nearly everything applies). The book is
1381 expensive, but really worth every penny to 1369 expensive, but really worth every penny to understand SMP locking.
1382 [ISBN: 0201633388] 1370 [ISBN: 0201633388]
1383 1371
1384 Thanks 1372 Thanks
1385 ====== 1373 ======
1386 1374
1387 Thanks to Telsa Gwynne for DocBooking, neaten 1375 Thanks to Telsa Gwynne for DocBooking, neatening and adding style.
1388 1376
1389 Thanks to Martin Pool, Philipp Rumpf, Stephen 1377 Thanks to Martin Pool, Philipp Rumpf, Stephen Rothwell, Paul Mackerras,
1390 Ruedi Aschwanden, Alan Cox, Manfred Spraul, T 1378 Ruedi Aschwanden, Alan Cox, Manfred Spraul, Tim Waugh, Pete Zaitcev,
1391 James Morris, Robert Love, Paul McKenney, Joh 1379 James Morris, Robert Love, Paul McKenney, John Ashby for proofreading,
1392 correcting, flaming, commenting. 1380 correcting, flaming, commenting.
1393 1381
1394 Thanks to the cabal for having no influence o 1382 Thanks to the cabal for having no influence on this document.
1395 1383
1396 Glossary 1384 Glossary
1397 ======== 1385 ========
1398 1386
1399 preemption 1387 preemption
1400 Prior to 2.5, or when ``CONFIG_PREEMPT`` is 1388 Prior to 2.5, or when ``CONFIG_PREEMPT`` is unset, processes in user
1401 context inside the kernel would not preempt 1389 context inside the kernel would not preempt each other (ie. you had that
1402 CPU until you gave it up, except for interr 1390 CPU until you gave it up, except for interrupts). With the addition of
1403 ``CONFIG_PREEMPT`` in 2.5.4, this changed: 1391 ``CONFIG_PREEMPT`` in 2.5.4, this changed: when in user context, higher
1404 priority tasks can "cut in": spinlocks were 1392 priority tasks can "cut in": spinlocks were changed to disable
1405 preemption, even on UP. 1393 preemption, even on UP.
1406 1394
1407 bh 1395 bh
1408 Bottom Half: for historical reasons, functi 1396 Bottom Half: for historical reasons, functions with '_bh' in them often
1409 now refer to any software interrupt, e.g. s 1397 now refer to any software interrupt, e.g. spin_lock_bh()
1410 blocks any software interrupt on the curren 1398 blocks any software interrupt on the current CPU. Bottom halves are
1411 deprecated, and will eventually be replaced 1399 deprecated, and will eventually be replaced by tasklets. Only one bottom
1412 half will be running at any time. 1400 half will be running at any time.
1413 1401
1414 Hardware Interrupt / Hardware IRQ 1402 Hardware Interrupt / Hardware IRQ
1415 Hardware interrupt request. in_hardirq() re 1403 Hardware interrupt request. in_hardirq() returns true in a
1416 hardware interrupt handler. 1404 hardware interrupt handler.
1417 1405
1418 Interrupt Context 1406 Interrupt Context
1419 Not user context: processing a hardware irq 1407 Not user context: processing a hardware irq or software irq. Indicated
1420 by the in_interrupt() macro returning true. 1408 by the in_interrupt() macro returning true.
1421 1409
1422 SMP 1410 SMP
1423 Symmetric Multi-Processor: kernels compiled 1411 Symmetric Multi-Processor: kernels compiled for multiple-CPU machines.
1424 (``CONFIG_SMP=y``). 1412 (``CONFIG_SMP=y``).
1425 1413
1426 Software Interrupt / softirq 1414 Software Interrupt / softirq
1427 Software interrupt handler. in_hardirq() re 1415 Software interrupt handler. in_hardirq() returns false;
1428 in_softirq() returns true. Tasklets and sof 1416 in_softirq() returns true. Tasklets and softirqs both
1429 fall into the category of 'software interru 1417 fall into the category of 'software interrupts'.
1430 1418
1431 Strictly speaking a softirq is one of up to 1419 Strictly speaking a softirq is one of up to 32 enumerated software
1432 interrupts which can run on multiple CPUs a 1420 interrupts which can run on multiple CPUs at once. Sometimes used to
1433 refer to tasklets as well (ie. all software 1421 refer to tasklets as well (ie. all software interrupts).
1434 1422
1435 tasklet 1423 tasklet
1436 A dynamically-registrable software interrup 1424 A dynamically-registrable software interrupt, which is guaranteed to
1437 only run on one CPU at a time. 1425 only run on one CPU at a time.
1438 1426
1439 timer 1427 timer
1440 A dynamically-registrable software interrup 1428 A dynamically-registrable software interrupt, which is run at (or close
1441 to) a given time. When running, it is just 1429 to) a given time. When running, it is just like a tasklet (in fact, they
1442 are called from the ``TIMER_SOFTIRQ``). 1430 are called from the ``TIMER_SOFTIRQ``).
1443 1431
1444 UP 1432 UP
1445 Uni-Processor: Non-SMP. (``CONFIG_SMP=n``). 1433 Uni-Processor: Non-SMP. (``CONFIG_SMP=n``).
1446 1434
1447 User Context 1435 User Context
1448 The kernel executing on behalf of a particu 1436 The kernel executing on behalf of a particular process (ie. a system
1449 call or trap) or kernel thread. You can tel 1437 call or trap) or kernel thread. You can tell which process with the
1450 ``current`` macro.) Not to be confused with 1438 ``current`` macro.) Not to be confused with userspace. Can be
1451 interrupted by software or hardware interru 1439 interrupted by software or hardware interrupts.
1452 1440
1453 Userspace 1441 Userspace
1454 A process executing its own code outside th 1442 A process executing its own code outside the kernel.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.