]>
Commit | Line | Data |
---|---|---|
8f0aff2a | 1 | .\" Page by b.hubert |
1abce893 MK |
2 | .\" and Copyright (C) 2015, Thomas Gleixner <tglx@linutronix.de> |
3 | .\" and Copyright (C) 2015, Michael Kerrisk <mtk.manpages@gmail.com> | |
2297bf0e | 4 | .\" |
2e46a6e7 | 5 | .\" %%%LICENSE_START(FREELY_REDISTRIBUTABLE) |
8f0aff2a | 6 | .\" may be freely modified and distributed |
8ff7380d | 7 | .\" %%%LICENSE_END |
fea681da MK |
8 | .\" |
9 | .\" Niki A. Rahimi (LTC Security Development, narahimi@us.ibm.com) | |
10 | .\" added ERRORS section. | |
11 | .\" | |
12 | .\" Modified 2004-06-17 mtk | |
13 | .\" Modified 2004-10-07 aeb, added FUTEX_REQUEUE, FUTEX_CMP_REQUEUE | |
14 | .\" | |
47f5c4ba | 15 | .\" FIXME Still to integrate are some points from Torvald Riegel's mail of |
9915ea23 | 16 | .\" 2015-01-23: |
47f5c4ba MK |
17 | .\" http://thread.gmane.org/gmane.linux.kernel/1703405/focus=7977 |
18 | .\" | |
78e85692 | 19 | .\" FIXME Do we need to add some text regarding Torvald Riegel's 2015-01-24 mail |
9915ea23 | 20 | .\" http://thread.gmane.org/gmane.linux.kernel/1703405/focus=1873242 |
02182e7c | 21 | .\" |
6e00b7a8 | 22 | .TH FUTEX 2 2021-08-27 "Linux" "Linux Programmer's Manual" |
fea681da | 23 | .SH NAME |
ce154705 | 24 | futex \- fast user-space locking |
fea681da | 25 | .SH SYNOPSIS |
9d9dc1e8 | 26 | .nf |
68e4db0a | 27 | .PP |
0aa385fe AC |
28 | .BR "#include <linux/futex.h>" " /* Definition of " FUTEX_* " constants */" |
29 | .BR "#include <sys/syscall.h>" " /* Definition of " SYS_* " constants */" | |
30 | .B #include <unistd.h> | |
68e4db0a | 31 | .PP |
0aa385fe AC |
32 | .BI "long syscall(SYS_futex, uint32_t *" uaddr ", int " futex_op \ |
33 | ", uint32_t " val , | |
34 | .BI " const struct timespec *" timeout , \ | |
9bfc9cb1 | 35 | " \fR /* or: \fBuint32_t \fIval2\fP */" |
0aa385fe | 36 | .BI " uint32_t *" uaddr2 ", uint32_t " val3 ); |
9d9dc1e8 | 37 | .fi |
dbfe9c70 | 38 | .PP |
b939d6e4 | 39 | .IR Note : |
0aa385fe AC |
40 | glibc provides no wrapper for |
41 | .BR futex (), | |
42 | necessitating the use of | |
43 | .BR syscall (2). | |
47297adb | 44 | .SH DESCRIPTION |
fea681da | 45 | The |
e511ffb6 | 46 | .BR futex () |
4b35dc5d | 47 | system call provides a method for waiting until a certain condition becomes |
077981d4 MK |
48 | true. |
49 | It is typically used as a blocking construct in the context of | |
d45f244c MK |
50 | shared-memory synchronization. |
51 | When using futexes, the majority of | |
52 | the synchronization operations are performed in user space. | |
bc54ed38 | 53 | A user-space program employs the |
d45f244c | 54 | .BR futex () |
ca4e5b2b | 55 | system call only when it is likely that the program has to block for |
4c8cb0ff | 56 | a longer time until the condition becomes true. |
bc54ed38 | 57 | Other |
d45f244c | 58 | .BR futex () |
bc54ed38 MK |
59 | operations can be used to wake any processes or threads waiting |
60 | for a particular condition. | |
efeece04 | 61 | .PP |
7e8dcabc MK |
62 | A futex is a 32-bit value\(emreferred to below as a |
63 | .IR "futex word" \(emwhose | |
64 | address is supplied to the | |
4b35dc5d | 65 | .BR futex () |
7e8dcabc | 66 | system call. |
c3f4c019 | 67 | (Futexes are 32 bits in size on all platforms, including 64-bit systems.) |
7e8dcabc MK |
68 | All futex operations are governed by this value. |
69 | In order to share a futex between processes, | |
70 | the futex is placed in a region of shared memory, | |
71 | created using (for example) | |
72 | .BR mmap (2) | |
73 | or | |
74 | .BR shmat (2). | |
c3f4c019 | 75 | (Thus, the futex word may have different |
7e8dcabc MK |
76 | virtual addresses in different processes, |
77 | but these addresses all refer to the same location in physical memory.) | |
ca4e5b2b MK |
78 | In a multithreaded program, it is sufficient to place the futex word |
79 | in a global variable shared by all threads. | |
efeece04 | 80 | .PP |
0c3ec26b MK |
81 | When executing a futex operation that requests to block a thread, |
82 | the kernel will block only if the futex word has the value that the | |
55f9e85e MK |
83 | calling thread supplied (as one of the arguments of the |
84 | .BR futex () | |
85 | call) as the expected value of the futex word. | |
9d32a39b MK |
86 | The loading of the futex word's value, |
87 | the comparison of that value with the expected value, | |
bc54ed38 | 88 | and the actual blocking will happen atomically and will be totally ordered |
da894b18 | 89 | with respect to concurrent operations performed by other threads |
0fb87d16 | 90 | on the same futex word. |
da894b18 MK |
91 | .\" Notes from Darren Hart (Dec 2015): |
92 | .\" Totally ordered with respect futex operations refers to semantics | |
93 | .\" of the ACQUIRE/RELEASE operations and how they impact ordering of | |
94 | .\" memory reads and writes. The kernel futex operations are protected | |
f6615c42 | 95 | .\" by spinlocks, which ensure that all operations are serialized |
da894b18 MK |
96 | .\" with respect to one another. |
97 | .\" | |
98 | .\" This is a lot to attempt to define in this document. Perhaps a | |
99 | .\" reference to linux/Documentation/memory-barriers.txt as a footnote | |
100 | .\" would be sufficient? Or perhaps for this manual, "serialized" would | |
101 | .\" be sufficient, with a footnote regarding "totally ordered" and a | |
102 | .\" pointer to the memory-barrier documentation? | |
b80daba2 | 103 | Thus, the futex word is used to connect the synchronization in user space |
9d32a39b | 104 | with the implementation of blocking by the kernel. |
55f9e85e | 105 | Analogously to an atomic |
4b35dc5d | 106 | compare-and-exchange operation that potentially changes shared memory, |
077981d4 | 107 | blocking via a futex is an atomic compare-and-block operation. |
d6bb5a38 | 108 | .\" FIXME(Torvald Riegel): |
61066e14 MK |
109 | .\" Eventually we want to have some text in NOTES to satisfy |
110 | .\" the reference in the following sentence | |
111 | .\" See NOTES for a detailed specification of | |
112 | .\" the synchronization semantics. | |
efeece04 | 113 | .PP |
ca4e5b2b | 114 | One use of futexes is for implementing locks. |
c0dc758e MK |
115 | The state of the lock (i.e., acquired or not acquired) |
116 | can be represented as an atomically accessed flag in shared memory. | |
4c8cb0ff | 117 | In the uncontended case, |
c3f4c019 | 118 | a thread can access or modify the lock state with atomic instructions, |
4c8cb0ff MK |
119 | for example atomically changing it from not acquired to acquired |
120 | using an atomic compare-and-exchange instruction. | |
55f9e85e MK |
121 | (Such instructions are performed entirely in user mode, |
122 | and the kernel maintains no information about the lock state.) | |
123 | On the other hand, a thread may be unable to acquire a lock because | |
8e754e12 | 124 | it is already acquired by another thread. |
55f9e85e | 125 | It then may pass the lock's flag as a futex word and the value |
0c3ec26b | 126 | representing the acquired state as the expected value to a |
8e754e12 HS |
127 | .BR futex () |
128 | wait operation. | |
55f9e85e | 129 | This |
8e754e12 | 130 | .BR futex () |
bc54ed38 | 131 | operation will block if and only if the lock is still acquired |
f6615c42 | 132 | (i.e., the value in the futex word still matches the "acquired state"). |
077981d4 | 133 | When releasing the lock, a thread has to first reset the |
0c3ec26b | 134 | lock state to not acquired and then execute a futex |
55f9e85e | 135 | operation that wakes threads blocked on the lock flag used as a futex word |
f6615c42 | 136 | (this can be further optimized to avoid unnecessary wake-ups). |
077981d4 | 137 | See |
4b35dc5d TR |
138 | .BR futex (7) |
139 | for more detail on how to use futexes. | |
efeece04 | 140 | .PP |
4b35dc5d | 141 | Besides the basic wait and wake-up futex functionality, there are further |
077981d4 | 142 | futex operations aimed at supporting more complex use cases. |
efeece04 | 143 | .PP |
ca4e5b2b | 144 | Note that |
2af84f99 | 145 | no explicit initialization or destruction is necessary to use futexes; |
4c8cb0ff MK |
146 | the kernel maintains a futex |
147 | (i.e., the kernel-internal implementation artifact) | |
4b35dc5d TR |
148 | only while operations such as |
149 | .BR FUTEX_WAIT , | |
150 | described below, are being performed on a particular futex word. | |
a663ca5a MK |
151 | .\" |
152 | .SS Arguments | |
fea681da MK |
153 | The |
154 | .I uaddr | |
077981d4 MK |
155 | argument points to the futex word. |
156 | On all platforms, futexes are four-byte | |
4b35dc5d | 157 | integers that must be aligned on a four-byte boundary. |
f388ba70 MK |
158 | The operation to perform on the futex is specified in the |
159 | .I futex_op | |
160 | argument; | |
161 | .IR val | |
162 | is a value whose meaning and purpose depends on | |
163 | .IR futex_op . | |
efeece04 | 164 | .PP |
36ab2074 MK |
165 | The remaining arguments |
166 | .RI ( timeout , | |
167 | .IR uaddr2 , | |
168 | and | |
169 | .IR val3 ) | |
170 | are required only for certain of the futex operations described below. | |
171 | Where one of these arguments is not required, it is ignored. | |
efeece04 | 172 | .PP |
36ab2074 MK |
173 | For several blocking operations, the |
174 | .I timeout | |
175 | argument is a pointer to a | |
176 | .IR timespec | |
177 | structure that specifies a timeout for the operation. | |
178 | However, notwithstanding the prototype shown above, for some operations, | |
eb4aa521 MK |
179 | the least significant four bytes of this argument are instead |
180 | used as an integer whose meaning is determined by the operation. | |
768d3c23 MK |
181 | For these operations, the kernel casts the |
182 | .I timeout | |
10022b8e HS |
183 | value first to |
184 | .IR "unsigned long", | |
185 | then to | |
c6dc40a2 | 186 | .IR uint32_t , |
768d3c23 MK |
187 | and in the remainder of this page, this argument is referred to as |
188 | .I val2 | |
189 | when interpreted in this fashion. | |
efeece04 | 190 | .PP |
de5a3bb4 | 191 | Where it is required, the |
36ab2074 | 192 | .IR uaddr2 |
4c8cb0ff MK |
193 | argument is a pointer to a second futex word that is employed |
194 | by the operation. | |
efeece04 | 195 | .PP |
36ab2074 MK |
196 | The interpretation of the final integer argument, |
197 | .IR val3 , | |
198 | depends on the operation. | |
a663ca5a MK |
199 | .\" |
200 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
201 | .\" | |
202 | .SS Futex operations | |
6be4bad7 | 203 | The |
d33602c4 | 204 | .I futex_op |
6be4bad7 MK |
205 | argument consists of two parts: |
206 | a command that specifies the operation to be performed, | |
5d771a4a | 207 | bitwise ORed with zero or more options that |
6be4bad7 | 208 | modify the behaviour of the operation. |
fc30eb79 | 209 | The options that may be included in |
d33602c4 | 210 | .I futex_op |
fc30eb79 TG |
211 | are as follows: |
212 | .TP | |
213 | .BR FUTEX_PRIVATE_FLAG " (since Linux 2.6.22)" | |
214 | .\" commit 34f01cc1f512fa783302982776895c73714ebbc2 | |
215 | This option bit can be employed with all futex operations. | |
e45f9735 | 216 | It tells the kernel that the futex is process-private and not shared |
0c3ec26b MK |
217 | with another process (i.e., it is being used for synchronization |
218 | only between threads of the same process). | |
943ccc52 MK |
219 | This allows the kernel to make some additional performance optimizations. |
220 | .\" I.e., It allows the kernel choose the fast path for validating | |
221 | .\" the user-space address and avoids expensive VMA lookups, | |
222 | .\" taking reference counts on file backing store, and so on. | |
efeece04 | 223 | .IP |
ae2c1774 | 224 | As a convenience, |
eeeee811 | 225 | .I <linux/futex.h> |
ae2c1774 | 226 | defines a set of constants with the suffix |
eeeee811 | 227 | .B _PRIVATE |
ae2c1774 | 228 | that are equivalents of all of the operations listed below, |
dcdfde26 | 229 | .\" except the obsolete FUTEX_FD, for which the "private" flag was |
ae2c1774 MK |
230 | .\" meaningless |
231 | but with the | |
232 | .BR FUTEX_PRIVATE_FLAG | |
233 | ORed into the constant value. | |
234 | Thus, there are | |
235 | .BR FUTEX_WAIT_PRIVATE , | |
236 | .BR FUTEX_WAKE_PRIVATE , | |
237 | and so on. | |
2e98bbc2 TG |
238 | .TP |
239 | .BR FUTEX_CLOCK_REALTIME " (since Linux 2.6.28)" | |
240 | .\" commit 1acdac104668a0834cfa267de9946fac7764d486 | |
4a7e5b05 | 241 | This option bit can be employed only with the |
949ceae3 MK |
242 | .BR FUTEX_WAIT_BITSET , |
243 | .BR FUTEX_WAIT_REQUEUE_PI , | |
949ceae3 MK |
244 | (since Linux 4.5) |
245 | .\" commit 337f13046ff03717a9e99675284a817527440a49 | |
e79977ae KK |
246 | .BR FUTEX_WAIT , |
247 | and | |
eeeee811 | 248 | (since Linux 5.14) |
e79977ae | 249 | .\" commit bf22a6976897977b0a3f1aeba6823c959fc4fdae |
eeeee811 | 250 | .B FUTEX_LOCK_PI2 |
c84cf68c | 251 | operations. |
efeece04 | 252 | .IP |
8064bfa5 | 253 | If this option is set, the kernel measures the |
f2103b26 | 254 | .I timeout |
8064bfa5 | 255 | against the |
eeeee811 | 256 | .B CLOCK_REALTIME |
8064bfa5 | 257 | clock. |
efeece04 | 258 | .IP |
8064bfa5 | 259 | If this option is not set, the kernel measures the |
f2103b26 | 260 | .I timeout |
8064bfa5 | 261 | against the |
eeeee811 | 262 | .B CLOCK_MONOTONIC |
1c952cf5 | 263 | clock. |
6be4bad7 MK |
264 | .PP |
265 | The operation specified in | |
d33602c4 | 266 | .I futex_op |
6be4bad7 | 267 | is one of the following: |
70b06b90 MK |
268 | .\" |
269 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
270 | .\" | |
fea681da | 271 | .TP |
81c9d87e MK |
272 | .BR FUTEX_WAIT " (since Linux 2.6.0)" |
273 | .\" Strictly speaking, since some time in 2.5.x | |
f065673c | 274 | This operation tests that the value at the |
4b35dc5d | 275 | futex word pointed to by the address |
fea681da | 276 | .I uaddr |
4b35dc5d | 277 | still contains the expected value |
fea681da | 278 | .IR val , |
fd105614 | 279 | and if so, then sleeps waiting for a |
682edefb | 280 | .B FUTEX_WAKE |
fd105614 | 281 | operation on the futex word. |
077981d4 | 282 | The load of the value of the futex word is an atomic memory |
4b35dc5d | 283 | access (i.e., using atomic machine instructions of the respective |
077981d4 MK |
284 | architecture). |
285 | This load, the comparison with the expected value, and | |
fd105614 | 286 | starting to sleep are performed atomically |
da56650a | 287 | .\" FIXME: Torvald, I think we may need to add some explanation of |
61066e14 | 288 | .\" "totally ordered" here. |
fd105614 MK |
289 | and totally ordered |
290 | with respect to other futex operations on the same futex word. | |
c0dc758e MK |
291 | If the thread starts to sleep, |
292 | it is considered a waiter on this futex word. | |
f065673c MK |
293 | If the futex value does not match |
294 | .IR val , | |
4710334a | 295 | then the call fails immediately with the error |
badbf70c | 296 | .BR EAGAIN . |
efeece04 | 297 | .IP |
4b35dc5d | 298 | The purpose of the comparison with the expected value is to prevent lost |
fd105614 MK |
299 | wake-ups. |
300 | If another thread changed the value of the futex word after the | |
c0dc758e MK |
301 | calling thread decided to block based on the prior value, |
302 | and if the other thread executed a | |
4b35dc5d TR |
303 | .BR FUTEX_WAKE |
304 | operation (or similar wake-up) after the value change and before this | |
f065673c | 305 | .BR FUTEX_WAIT |
bc54ed38 MK |
306 | operation, then the calling thread will observe the |
307 | value change and will not start to sleep. | |
efeece04 | 308 | .IP |
c13182ef | 309 | If the |
fea681da | 310 | .I timeout |
40d2dab9 | 311 | is not NULL, the structure it points to specifies a |
40d2dab9 | 312 | timeout for the wait. |
ac991a11 MK |
313 | (This interval will be rounded up to the system clock granularity, |
314 | and is guaranteed not to expire early.) | |
a6918f1d | 315 | The timeout is by default measured according to the |
1c952cf5 | 316 | .BR CLOCK_MONOTONIC |
a01c3098 MK |
317 | clock, but, since Linux 4.5, the |
318 | .BR CLOCK_REALTIME | |
319 | clock can be selected by specifying | |
320 | .BR FUTEX_CLOCK_REALTIME | |
321 | in | |
322 | .IR futex_op . | |
82a6092b MK |
323 | If |
324 | .I timeout | |
325 | is NULL, the call blocks indefinitely. | |
efeece04 | 326 | .IP |
4100abc5 MK |
327 | .IR Note : |
328 | for | |
329 | .BR FUTEX_WAIT , | |
330 | .IR timeout | |
331 | is interpreted as a | |
332 | .IR relative | |
333 | value. | |
334 | This differs from other futex operations, where | |
335 | .I timeout | |
336 | is interpreted as an absolute value. | |
337 | To obtain the equivalent of | |
338 | .BR FUTEX_WAIT | |
339 | with an absolute timeout, employ | |
340 | .BR FUTEX_WAIT_BITSET | |
341 | with | |
342 | .IR val3 | |
343 | specified as | |
344 | .BR FUTEX_BITSET_MATCH_ANY . | |
efeece04 | 345 | .IP |
c13182ef | 346 | The arguments |
fea681da MK |
347 | .I uaddr2 |
348 | and | |
349 | .I val3 | |
350 | are ignored. | |
9915ea23 MK |
351 | .\" FIXME . (Torvald) I think we should remove this. Or maybe adapt to a |
352 | .\" different example. | |
353 | .\" | |
354 | .\" For | |
355 | .\" .BR futex (7), | |
356 | .\" this call is executed if decrementing the count gave a negative value | |
357 | .\" (indicating contention), | |
358 | .\" and will sleep until another process or thread releases | |
359 | .\" the futex and executes the | |
360 | .\" .B FUTEX_WAKE | |
361 | .\" operation. | |
70b06b90 MK |
362 | .\" |
363 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
364 | .\" | |
fea681da | 365 | .TP |
81c9d87e MK |
366 | .BR FUTEX_WAKE " (since Linux 2.6.0)" |
367 | .\" Strictly speaking, since Linux 2.5.x | |
f065673c MK |
368 | This operation wakes at most |
369 | .I val | |
4b35dc5d | 370 | of the waiters that are waiting (e.g., inside |
f065673c | 371 | .BR FUTEX_WAIT ) |
4b35dc5d | 372 | on the futex word at the address |
f065673c MK |
373 | .IR uaddr . |
374 | Most commonly, | |
375 | .I val | |
376 | is specified as either 1 (wake up a single waiter) or | |
377 | .BR INT_MAX | |
378 | (wake up all waiters). | |
730bfbda MK |
379 | No guarantee is provided about which waiters are awoken |
380 | (e.g., a waiter with a higher scheduling priority is not guaranteed | |
381 | to be awoken in preference to a waiter with a lower priority). | |
efeece04 | 382 | .IP |
fea681da MK |
383 | The arguments |
384 | .IR timeout , | |
c8b921bd | 385 | .IR uaddr2 , |
fea681da MK |
386 | and |
387 | .I val3 | |
388 | are ignored. | |
9915ea23 MK |
389 | .\" FIXME . (Torvald) I think we should remove this. Or maybe adapt to |
390 | .\" a different example. | |
391 | .\" | |
4c8cb0ff MK |
392 | .\" For |
393 | .\" .BR futex (7), | |
394 | .\" this is executed if incrementing the count showed that | |
395 | .\" there were waiters, | |
396 | .\" once the futex value has been set to 1 | |
397 | .\" (indicating that it is available). | |
398 | .\" | |
9915ea23 | 399 | .\" How does "incrementing the count show that there were waiters"? |
70b06b90 MK |
400 | .\" |
401 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
402 | .\" | |
a7c2bf45 MK |
403 | .TP |
404 | .BR FUTEX_FD " (from Linux 2.6.0 up to and including Linux 2.6.25)" | |
405 | .\" Strictly speaking, from Linux 2.5.x to 2.6.25 | |
4c8cb0ff MK |
406 | This operation creates a file descriptor that is associated with |
407 | the futex at | |
a7c2bf45 | 408 | .IR uaddr . |
bdc5957a MK |
409 | The caller must close the returned file descriptor after use. |
410 | When another process or thread performs a | |
a7c2bf45 | 411 | .BR FUTEX_WAKE |
4b35dc5d | 412 | on the futex word, the file descriptor indicates as being readable with |
a7c2bf45 MK |
413 | .BR select (2), |
414 | .BR poll (2), | |
415 | and | |
416 | .BR epoll (7) | |
efeece04 | 417 | .IP |
f1d2171d | 418 | The file descriptor can be used to obtain asynchronous notifications: if |
a7c2bf45 | 419 | .I val |
ca4e5b2b | 420 | is nonzero, then, when another process or thread executes a |
a7c2bf45 MK |
421 | .BR FUTEX_WAKE , |
422 | the caller will receive the signal number that was passed in | |
423 | .IR val . | |
efeece04 | 424 | .IP |
a7c2bf45 MK |
425 | The arguments |
426 | .IR timeout , | |
d556548b | 427 | .IR uaddr2 , |
a7c2bf45 MK |
428 | and |
429 | .I val3 | |
430 | are ignored. | |
efeece04 | 431 | .IP |
a7c2bf45 MK |
432 | Because it was inherently racy, |
433 | .B FUTEX_FD | |
434 | has been removed | |
435 | .\" commit 82af7aca56c67061420d618cc5a30f0fd4106b80 | |
436 | from Linux 2.6.26 onward. | |
70b06b90 MK |
437 | .\" |
438 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
439 | .\" | |
a7c2bf45 MK |
440 | .TP |
441 | .BR FUTEX_REQUEUE " (since Linux 2.6.0)" | |
a7c2bf45 | 442 | This operation performs the same task as |
27dd3a6e MK |
443 | .BR FUTEX_CMP_REQUEUE |
444 | (see below), except that no check is made using the value in | |
a7c2bf45 MK |
445 | .IR val3 . |
446 | (The argument | |
447 | .I val3 | |
448 | is ignored.) | |
70b06b90 MK |
449 | .\" |
450 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
451 | .\" | |
a7c2bf45 MK |
452 | .TP |
453 | .BR FUTEX_CMP_REQUEUE " (since Linux 2.6.7)" | |
4b35dc5d | 454 | This operation first checks whether the location |
a7c2bf45 MK |
455 | .I uaddr |
456 | still contains the value | |
457 | .IR val3 . | |
458 | If not, the operation fails with the error | |
459 | .BR EAGAIN . | |
4b35dc5d | 460 | Otherwise, the operation wakes up a maximum of |
a7c2bf45 MK |
461 | .I val |
462 | waiters that are waiting on the futex at | |
463 | .IR uaddr . | |
464 | If there are more than | |
465 | .I val | |
466 | waiters, then the remaining waiters are removed | |
467 | from the wait queue of the source futex at | |
468 | .I uaddr | |
469 | and added to the wait queue of the target futex at | |
470 | .IR uaddr2 . | |
471 | The | |
768d3c23 | 472 | .I val2 |
936876a9 | 473 | argument specifies an upper limit on the number of waiters |
a7c2bf45 | 474 | that are requeued to the futex at |
768d3c23 | 475 | .IR uaddr2 . |
efeece04 | 476 | .IP |
d6bb5a38 MK |
477 | .\" FIXME(Torvald) Is the following correct? Or is just the decision |
478 | .\" which threads to wake or requeue part of the atomic operation? | |
4b35dc5d TR |
479 | The load from |
480 | .I uaddr | |
4c8cb0ff MK |
481 | is an atomic memory access (i.e., using atomic machine instructions of |
482 | the respective architecture). | |
077981d4 | 483 | This load, the comparison with |
4b35dc5d | 484 | .IR val3 , |
4c8cb0ff MK |
485 | and the requeueing of any waiters are performed atomically and totally |
486 | ordered with respect to other operations on the same futex word. | |
ee65b0e8 MK |
487 | .\" Notes from a f2f conversation with Thomas Gleixner (Aug 2015): ### |
488 | .\" The operation is serialized with respect to operations on both | |
489 | .\" source and target futex. No other waiter can enqueue itself | |
490 | .\" for waiting and no other waiter can dequeue itself because of | |
491 | .\" a timeout or signal. | |
efeece04 | 492 | .IP |
a7c2bf45 MK |
493 | Typical values to specify for |
494 | .I val | |
ed1819cf | 495 | are 0 or 1. |
a7c2bf45 MK |
496 | (Specifying |
497 | .BR INT_MAX | |
498 | is not useful, because it would make the | |
499 | .BR FUTEX_CMP_REQUEUE | |
500 | operation equivalent to | |
501 | .BR FUTEX_WAKE .) | |
936876a9 | 502 | The limit value specified via |
768d3c23 MK |
503 | .I val2 |
504 | is typically either 1 or | |
a7c2bf45 MK |
505 | .BR INT_MAX . |
506 | (Specifying the argument as 0 is not useful, because it would make the | |
507 | .BR FUTEX_CMP_REQUEUE | |
508 | operation equivalent to | |
509 | .BR FUTEX_WAIT .) | |
efeece04 | 510 | .IP |
627b50ce MK |
511 | The |
512 | .B FUTEX_CMP_REQUEUE | |
513 | operation was added as a replacement for the earlier | |
514 | .BR FUTEX_REQUEUE . | |
515 | The difference is that the check of the value at | |
516 | .I uaddr | |
517 | can be used to ensure that requeueing happens only under certain | |
518 | conditions, which allows race conditions to be avoided in certain use cases. | |
dcb410c3 | 519 | .\" But, as Rich Felker points out, there remain valid use cases for |
627b50ce MK |
520 | .\" FUTEX_REQUEUE, for example, when the calling thread is requeuing |
521 | .\" the target(s) to a lock that the calling thread owns | |
522 | .\" From: Rich Felker <dalias@libc.org> | |
523 | .\" Date: Wed, 29 Oct 2014 22:43:17 -0400 | |
524 | .\" To: Darren Hart <dvhart@infradead.org> | |
525 | .\" CC: libc-alpha@sourceware.org, ... | |
526 | .\" Subject: Re: Add futex wrapper to glibc? | |
efeece04 | 527 | .IP |
627b50ce MK |
528 | Both |
529 | .BR FUTEX_REQUEUE | |
530 | and | |
531 | .BR FUTEX_CMP_REQUEUE | |
532 | can be used to avoid "thundering herd" wake-ups that could occur when using | |
533 | .B FUTEX_WAKE | |
534 | in cases where all of the waiters that are woken need to acquire | |
535 | another futex. | |
536 | Consider the following scenario, | |
537 | where multiple waiter threads are waiting on B, | |
538 | a wait queue implemented using a futex: | |
efeece04 | 539 | .IP |
627b50ce | 540 | .in +4n |
b76974c1 | 541 | .EX |
627b50ce MK |
542 | lock(A) |
543 | while (!check_value(V)) { | |
544 | unlock(A); | |
545 | block_on(B); | |
546 | lock(A); | |
547 | }; | |
548 | unlock(A); | |
b76974c1 | 549 | .EE |
627b50ce | 550 | .in |
efeece04 | 551 | .IP |
627b50ce MK |
552 | If a waker thread used |
553 | .BR FUTEX_WAKE , | |
554 | then all waiters waiting on B would be woken up, | |
67c67ff2 | 555 | and they would all try to acquire lock A. |
627b50ce MK |
556 | However, waking all of the threads in this manner would be pointless because |
557 | all except one of the threads would immediately block on lock A again. | |
558 | By contrast, a requeue operation wakes just one waiter and moves | |
559 | the other waiters to lock A, | |
560 | and when the woken waiter unlocks A then the next waiter can proceed. | |
43d16602 | 561 | .\" |
70b06b90 MK |
562 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" |
563 | .\" | |
fea681da | 564 | .TP |
d67e21f5 MK |
565 | .BR FUTEX_WAKE_OP " (since Linux 2.6.14)" |
566 | .\" commit 4732efbeb997189d9f9b04708dc26bf8613ed721 | |
6bac3b85 MK |
567 | .\" Author: Jakub Jelinek <jakub@redhat.com> |
568 | .\" Date: Tue Sep 6 15:16:25 2005 -0700 | |
9915ea23 | 569 | .\" FIXME. (Torvald) The glibc condvar implementation is currently being |
4c8cb0ff MK |
570 | .\" revised (e.g., to not use an internal lock anymore). |
571 | .\" It is probably more future-proof to remove this paragraph. | |
d6bb5a38 | 572 | .\" [Torvald, do you have an update here?] |
6bac3b85 MK |
573 | This operation was added to support some user-space use cases |
574 | where more than one futex must be handled at the same time. | |
575 | The most notable example is the implementation of | |
576 | .BR pthread_cond_signal (3), | |
577 | which requires operations on two futexes, | |
578 | the one used to implement the mutex and the one used in the implementation | |
579 | of the wait queue associated with the condition variable. | |
580 | .BR FUTEX_WAKE_OP | |
581 | allows such cases to be implemented without leading to | |
582 | high rates of contention and context switching. | |
efeece04 | 583 | .IP |
6bac3b85 | 584 | The |
57f2d48b | 585 | .BR FUTEX_WAKE_OP |
e61abc20 | 586 | operation is equivalent to executing the following code atomically |
4c8cb0ff MK |
587 | and totally ordered with respect to other futex operations on |
588 | any of the two supplied futex words: | |
efeece04 | 589 | .IP |
6bac3b85 | 590 | .in +4n |
b76974c1 | 591 | .EX |
2253ecf0 AC |
592 | uint32_t oldval = *(uint32_t *) uaddr2; |
593 | *(uint32_t *) uaddr2 = oldval \fIop\fP \fIoparg\fP; | |
6bac3b85 MK |
594 | futex(uaddr, FUTEX_WAKE, val, 0, 0, 0); |
595 | if (oldval \fIcmp\fP \fIcmparg\fP) | |
768d3c23 | 596 | futex(uaddr2, FUTEX_WAKE, val2, 0, 0, 0); |
b76974c1 | 597 | .EE |
6bac3b85 | 598 | .in |
efeece04 | 599 | .IP |
6bac3b85 | 600 | In other words, |
57f2d48b | 601 | .BR FUTEX_WAKE_OP |
6bac3b85 MK |
602 | does the following: |
603 | .RS | |
604 | .IP * 3 | |
4b35dc5d TR |
605 | saves the original value of the futex word at |
606 | .IR uaddr2 | |
607 | and performs an operation to modify the value of the futex at | |
6bac3b85 | 608 | .IR uaddr2 ; |
4c8cb0ff MK |
609 | this is an atomic read-modify-write memory access (i.e., using atomic |
610 | machine instructions of the respective architecture) | |
6bac3b85 MK |
611 | .IP * |
612 | wakes up a maximum of | |
613 | .I val | |
4b35dc5d | 614 | waiters on the futex for the futex word at |
6bac3b85 MK |
615 | .IR uaddr ; |
616 | and | |
617 | .IP * | |
4c8cb0ff MK |
618 | dependent on the results of a test of the original value of the |
619 | futex word at | |
6bac3b85 MK |
620 | .IR uaddr2 , |
621 | wakes up a maximum of | |
768d3c23 | 622 | .I val2 |
4b35dc5d | 623 | waiters on the futex for the futex word at |
6bac3b85 MK |
624 | .IR uaddr2 . |
625 | .RE | |
626 | .IP | |
6bac3b85 MK |
627 | The operation and comparison that are to be performed are encoded |
628 | in the bits of the argument | |
629 | .IR val3 . | |
630 | Pictorially, the encoding is: | |
efeece04 | 631 | .IP |
3cf61490 | 632 | .in +4n |
b76974c1 | 633 | .EX |
f6af90e7 MK |
634 | +---+---+-----------+-----------+ |
635 | |op |cmp| oparg | cmparg | | |
636 | +---+---+-----------+-----------+ | |
637 | 4 4 12 12 <== # of bits | |
b76974c1 | 638 | .EE |
6bac3b85 | 639 | .in |
efeece04 | 640 | .IP |
6bac3b85 | 641 | Expressed in code, the encoding is: |
efeece04 | 642 | .IP |
6bac3b85 | 643 | .in +4n |
b76974c1 | 644 | .EX |
d1a71985 MK |
645 | #define FUTEX_OP(op, oparg, cmp, cmparg) \e |
646 | (((op & 0xf) << 28) | \e | |
647 | ((cmp & 0xf) << 24) | \e | |
648 | ((oparg & 0xfff) << 12) | \e | |
6bac3b85 | 649 | (cmparg & 0xfff)) |
b76974c1 | 650 | .EE |
6bac3b85 | 651 | .in |
efeece04 | 652 | .IP |
6bac3b85 MK |
653 | In the above, |
654 | .I op | |
655 | and | |
656 | .I cmp | |
657 | are each one of the codes listed below. | |
658 | The | |
659 | .I oparg | |
660 | and | |
661 | .I cmparg | |
662 | components are literal numeric values, except as noted below. | |
efeece04 | 663 | .IP |
6bac3b85 MK |
664 | The |
665 | .I op | |
666 | component has one of the following values: | |
efeece04 | 667 | .IP |
6bac3b85 | 668 | .in +4n |
b76974c1 | 669 | .EX |
6bac3b85 MK |
670 | FUTEX_OP_SET 0 /* uaddr2 = oparg; */ |
671 | FUTEX_OP_ADD 1 /* uaddr2 += oparg; */ | |
672 | FUTEX_OP_OR 2 /* uaddr2 |= oparg; */ | |
af2d18b2 | 673 | FUTEX_OP_ANDN 3 /* uaddr2 &= \(tioparg; */ |
9ca13180 | 674 | FUTEX_OP_XOR 4 /* uaddr2 \(ha= oparg; */ |
b76974c1 | 675 | .EE |
6bac3b85 | 676 | .in |
efeece04 | 677 | .IP |
5d771a4a | 678 | In addition, bitwise ORing the following value into |
6bac3b85 MK |
679 | .I op |
680 | causes | |
681 | .IR "(1\ <<\ oparg)" | |
682 | to be used as the operand: | |
efeece04 | 683 | .IP |
6bac3b85 | 684 | .in +4n |
b76974c1 | 685 | .EX |
6bac3b85 | 686 | FUTEX_OP_ARG_SHIFT 8 /* Use (1 << oparg) as operand */ |
b76974c1 | 687 | .EE |
6bac3b85 | 688 | .in |
efeece04 | 689 | .IP |
6bac3b85 MK |
690 | The |
691 | .I cmp | |
692 | field is one of the following: | |
efeece04 | 693 | .IP |
6bac3b85 | 694 | .in +4n |
b76974c1 | 695 | .EX |
6bac3b85 MK |
696 | FUTEX_OP_CMP_EQ 0 /* if (oldval == cmparg) wake */ |
697 | FUTEX_OP_CMP_NE 1 /* if (oldval != cmparg) wake */ | |
698 | FUTEX_OP_CMP_LT 2 /* if (oldval < cmparg) wake */ | |
699 | FUTEX_OP_CMP_LE 3 /* if (oldval <= cmparg) wake */ | |
700 | FUTEX_OP_CMP_GT 4 /* if (oldval > cmparg) wake */ | |
701 | FUTEX_OP_CMP_GE 5 /* if (oldval >= cmparg) wake */ | |
b76974c1 | 702 | .EE |
6bac3b85 | 703 | .in |
efeece04 | 704 | .IP |
6bac3b85 MK |
705 | The return value of |
706 | .BR FUTEX_WAKE_OP | |
707 | is the sum of the number of waiters woken on the futex | |
708 | .IR uaddr | |
709 | plus the number of waiters woken on the futex | |
710 | .IR uaddr2 . | |
70b06b90 MK |
711 | .\" |
712 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
713 | .\" | |
d67e21f5 | 714 | .TP |
79c9b436 TG |
715 | .BR FUTEX_WAIT_BITSET " (since Linux 2.6.25)" |
716 | .\" commit cd689985cf49f6ff5c8eddc48d98b9d581d9475d | |
fd9e59d4 | 717 | This operation is like |
79c9b436 TG |
718 | .BR FUTEX_WAIT |
719 | except that | |
720 | .I val3 | |
84abf4ba | 721 | is used to provide a 32-bit bit mask to the kernel. |
2ae96e8a | 722 | This bit mask, in which at least one bit must be set, |
6c38ce7f | 723 | is stored in the kernel-internal state of the waiter. |
79c9b436 TG |
724 | See the description of |
725 | .BR FUTEX_WAKE_BITSET | |
726 | for further details. | |
efeece04 | 727 | .IP |
8064bfa5 MK |
728 | If |
729 | .I timeout | |
730 | is not NULL, the structure it points to specifies | |
731 | an absolute timeout for the wait operation. | |
732 | If | |
733 | .I timeout | |
734 | is NULL, the operation can block indefinitely. | |
efeece04 | 735 | .IP |
79c9b436 TG |
736 | The |
737 | .I uaddr2 | |
738 | argument is ignored. | |
70b06b90 MK |
739 | .\" |
740 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
741 | .\" | |
79c9b436 | 742 | .TP |
d67e21f5 MK |
743 | .BR FUTEX_WAKE_BITSET " (since Linux 2.6.25)" |
744 | .\" commit cd689985cf49f6ff5c8eddc48d98b9d581d9475d | |
55cc422d TG |
745 | This operation is the same as |
746 | .BR FUTEX_WAKE | |
747 | except that the | |
e24fbf10 | 748 | .I val3 |
5e1456d4 | 749 | argument is used to provide a 32-bit bit mask to the kernel. |
6c38ce7f MK |
750 | This bit mask, in which at least one bit must be set, |
751 | is used to select which waiters should be woken up. | |
5d771a4a | 752 | The selection is done by a bitwise AND of the "wake" bit mask |
98d769c0 MK |
753 | (i.e., the value in |
754 | .IR val3 ) | |
5e1456d4 MK |
755 | and the bit mask which is stored in the kernel-internal |
756 | state of the waiter (the "wait" bit mask that is set using | |
98d769c0 MK |
757 | .BR FUTEX_WAIT_BITSET ). |
758 | All of the waiters for which the result of the AND is nonzero are woken up; | |
759 | the remaining waiters are left sleeping. | |
efeece04 | 760 | .IP |
e9d4496b MK |
761 | The effect of |
762 | .BR FUTEX_WAIT_BITSET | |
763 | and | |
764 | .BR FUTEX_WAKE_BITSET | |
9732dd8b MK |
765 | is to allow selective wake-ups among multiple waiters that are blocked |
766 | on the same futex. | |
ac894879 | 767 | However, note that, depending on the use case, |
5e1456d4 | 768 | employing this bit-mask multiplexing feature on a |
ac894879 | 769 | futex can be less efficient than simply using multiple futexes, |
5e1456d4 | 770 | because employing bit-mask multiplexing requires the kernel |
e9d4496b MK |
771 | to check all waiters on a futex, |
772 | including those that are not interested in being woken up | |
5e1456d4 | 773 | (i.e., they do not have the relevant bit set in their "wait" bit mask). |
e9d4496b MK |
774 | .\" According to http://locklessinc.com/articles/futex_cheat_sheet/: |
775 | .\" | |
776 | .\" "The original reason for the addition of these extensions | |
777 | .\" was to improve the performance of pthread read-write locks | |
778 | .\" in glibc. However, the pthreads library no longer uses the | |
779 | .\" same locking algorithm, and these extensions are not used | |
780 | .\" without the bitset parameter being all ones. | |
e24fbf10 | 781 | .\" |
e9d4496b | 782 | .\" The page goes on to note that the FUTEX_WAIT_BITSET operation |
5e1456d4 | 783 | .\" is nevertheless used (with a bit mask of all ones) in order to |
e9d4496b MK |
784 | .\" obtain the absolute timeout functionality that is useful |
785 | .\" for efficiently implementing Pthreads APIs (which use absolute | |
786 | .\" timeouts); FUTEX_WAIT provides only relative timeouts. | |
efeece04 | 787 | .IP |
678c9986 MK |
788 | The constant |
789 | .BR FUTEX_BITSET_MATCH_ANY , | |
790 | which corresponds to all 32 bits set in the bit mask, can be used as the | |
791 | .I val3 | |
792 | argument for | |
793 | .BR FUTEX_WAIT_BITSET | |
98d769c0 | 794 | and |
678c9986 MK |
795 | .BR FUTEX_WAKE_BITSET . |
796 | Other than differences in the handling of the | |
98d769c0 | 797 | .I timeout |
678c9986 | 798 | argument, the |
9732dd8b | 799 | .BR FUTEX_WAIT |
678c9986 | 800 | operation is equivalent to |
9732dd8b | 801 | .BR FUTEX_WAIT_BITSET |
678c9986 MK |
802 | with |
803 | .IR val3 | |
804 | specified as | |
805 | .BR FUTEX_BITSET_MATCH_ANY ; | |
806 | that is, allow a wake-up by any waker. | |
807 | The | |
808 | .BR FUTEX_WAKE | |
809 | operation is equivalent to | |
9732dd8b | 810 | .BR FUTEX_WAKE_BITSET |
678c9986 MK |
811 | with |
812 | .IR val3 | |
813 | specified as | |
814 | .BR FUTEX_BITSET_MATCH_ANY ; | |
815 | that is, wake up any waiter(s). | |
efeece04 | 816 | .IP |
678c9986 MK |
817 | The |
818 | .I uaddr2 | |
819 | and | |
820 | .I timeout | |
821 | arguments are ignored. | |
bd90a5f9 | 822 | .\" |
70b06b90 | 823 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" |
bd90a5f9 MK |
824 | .\" |
825 | .SS Priority-inheritance futexes | |
b52e1cd4 MK |
826 | Linux supports priority-inheritance (PI) futexes in order to handle |
827 | priority-inversion problems that can be encountered with | |
828 | normal futex locks. | |
b565548b | 829 | Priority inversion is the problem that occurs when a high-priority |
bdc5957a MK |
830 | task is blocked waiting to acquire a lock held by a low-priority task, |
831 | while tasks at an intermediate priority continuously preempt | |
832 | the low-priority task from the CPU. | |
833 | Consequently, the low-priority task makes no progress toward | |
834 | releasing the lock, and the high-priority task remains blocked. | |
efeece04 | 835 | .PP |
7d20efd7 MK |
836 | Priority inheritance is a mechanism for dealing with |
837 | the priority-inversion problem. | |
bdc5957a MK |
838 | With this mechanism, when a high-priority task becomes blocked |
839 | by a lock held by a low-priority task, | |
9cee832c MK |
840 | the priority of the low-priority task is temporarily raised |
841 | to that of the high-priority task, | |
bdc5957a | 842 | so that it is not preempted by any intermediate level tasks, |
7d20efd7 MK |
843 | and can thus make progress toward releasing the lock. |
844 | To be effective, priority inheritance must be transitive, | |
bdc5957a | 845 | meaning that if a high-priority task blocks on a lock |
ca4e5b2b | 846 | held by a lower-priority task that is itself blocked by a lock |
bdc5957a | 847 | held by another intermediate-priority task |
7d20efd7 | 848 | (and so on, for chains of arbitrary length), |
b0f35fbb | 849 | then both of those tasks |
bdc5957a MK |
850 | (or more generally, all of the tasks in a lock chain) |
851 | have their priorities raised to be the same as the high-priority task. | |
efeece04 | 852 | .PP |
9e2b90ee | 853 | From a user-space perspective, |
39e9b2e1 MK |
854 | what makes a futex PI-aware is a policy agreement (described below) |
855 | between user space and the kernel about the value of the futex word, | |
601399f3 MK |
856 | coupled with the use of the PI-futex operations described below. |
857 | (Unlike the other futex operations described above, | |
858 | the PI-futex operations are designed | |
859 | for the implementation of very specific IPC mechanisms.) | |
860 | .\" | |
9e2b90ee MK |
861 | .\" Quoting Darren Hart: |
862 | .\" These opcodes paired with the PI futex value policy (described below) | |
863 | .\" defines a "futex" as PI aware. These were created very specifically | |
864 | .\" in support of PI pthread_mutexes, so it makes a lot more sense to | |
865 | .\" talk about a PI aware pthread_mutex, than a PI aware futex, since | |
866 | .\" there is a lot of policy and scaffolding that has to be built up | |
867 | .\" around it to use it properly (this is what a PI pthread_mutex is). | |
efeece04 | 868 | .PP |
ac894879 | 869 | .\" mtk: The following text is drawn from the Hart/Guniguntala paper |
1af427a4 | 870 | .\" (listed in SEE ALSO), but I have reworded some pieces |
8d825152 | 871 | .\" significantly. |
79d918c7 | 872 | .\" |
f0a9e8f4 | 873 | The PI-futex operations described below differ from the other |
4b35dc5d TR |
874 | futex operations in that they impose policy on the use of the value of the |
875 | futex word: | |
79d918c7 | 876 | .IP * 3 |
4b35dc5d | 877 | If the lock is not acquired, the futex word's value shall be 0. |
79d918c7 | 878 | .IP * |
4c8cb0ff MK |
879 | If the lock is acquired, the futex word's value shall |
880 | be the thread ID (TID; | |
4b35dc5d | 881 | see |
79d918c7 MK |
882 | .BR gettid (2)) |
883 | of the owning thread. | |
884 | .IP * | |
79d918c7 MK |
885 | If the lock is owned and there are threads contending for the lock, |
886 | then the | |
887 | .B FUTEX_WAITERS | |
4b35dc5d | 888 | bit shall be set in the futex word's value; in other words, this value is: |
efeece04 | 889 | .IP |
79d918c7 | 890 | FUTEX_WAITERS | TID |
601399f3 MK |
891 | .IP |
892 | (Note that is invalid for a PI futex word to have no owner and | |
893 | .BR FUTEX_WAITERS | |
894 | set.) | |
79d918c7 MK |
895 | .PP |
896 | With this policy in place, | |
fd105614 | 897 | a user-space application can acquire an unacquired |
601399f3 | 898 | lock or release a lock using atomic instructions executed in user mode |
fd105614 | 899 | (e.g., a compare-and-swap operation such as |
b52e1cd4 MK |
900 | .I cmpxchg |
901 | on the x86 architecture). | |
4c8cb0ff MK |
902 | Acquiring a lock simply consists of using compare-and-swap to atomically |
903 | set the futex word's value to the caller's TID if its previous value was 0. | |
4b35dc5d TR |
904 | Releasing a lock requires using compare-and-swap to set the futex word's |
905 | value to 0 if the previous value was the expected TID. | |
efeece04 | 906 | .PP |
4b35dc5d | 907 | If a futex is already acquired (i.e., has a nonzero value), |
b52e1cd4 | 908 | waiters must employ the |
79d918c7 | 909 | .B FUTEX_LOCK_PI |
e79977ae KK |
910 | or |
911 | .B FUTEX_LOCK_PI2 | |
912 | operations to acquire the lock. | |
4b35dc5d | 913 | If other threads are waiting for the lock, then the |
79d918c7 | 914 | .B FUTEX_WAITERS |
4c8cb0ff MK |
915 | bit is set in the futex value; |
916 | in this case, the lock owner must employ the | |
79d918c7 | 917 | .B FUTEX_UNLOCK_PI |
b52e1cd4 | 918 | operation to release the lock. |
efeece04 | 919 | .PP |
79d918c7 MK |
920 | In the cases where callers are forced into the kernel |
921 | (i.e., required to perform a | |
922 | .BR futex () | |
0c3ec26b | 923 | call), |
79d918c7 MK |
924 | they then deal directly with a so-called RT-mutex, |
925 | a kernel locking mechanism which implements the required | |
926 | priority-inheritance semantics. | |
927 | After the RT-mutex is acquired, the futex value is updated accordingly, | |
928 | before the calling thread returns to user space. | |
efeece04 | 929 | .PP |
a59fca75 | 930 | It is important to note |
ac894879 | 931 | .\" tglx (July 2015): |
30239c10 MK |
932 | .\" If there are multiple waiters on a pi futex then a wake pi operation |
933 | .\" will wake the first waiter and hand over the lock to this waiter. This | |
934 | .\" includes handing over the rtmutex which represents the futex in the | |
935 | .\" kernel. The strict requirement is that the futex owner and the rtmutex | |
936 | .\" owner must be the same, except for the update period which is | |
937 | .\" serialized by the futex internal locking. That means the kernel must | |
1d09c150 | 938 | .\" update the user-space value prior to returning to user space |
4b35dc5d | 939 | that the kernel will update the futex word's value prior |
79d918c7 | 940 | to returning to user space. |
601399f3 MK |
941 | (This prevents the possibility of the futex word's value ending |
942 | up in an invalid state, such as having an owner but the value being 0, | |
943 | or having waiters but not having the | |
944 | .B FUTEX_WAITERS | |
945 | bit set.) | |
efeece04 | 946 | .PP |
30239c10 MK |
947 | If a futex has an associated RT-mutex in the kernel |
948 | (i.e., there are blocked waiters) | |
949 | and the owner of the futex/RT-mutex dies unexpectedly, | |
950 | then the kernel cleans up the RT-mutex and hands it over to the next waiter. | |
951 | This in turn requires that the user-space value is updated accordingly. | |
952 | To indicate that this is required, the kernel sets the | |
953 | .B FUTEX_OWNER_DIED | |
954 | bit in the futex word along with the thread ID of the new owner. | |
8adaf0a7 MK |
955 | User space can detect this situation via the presence of the |
956 | .B FUTEX_OWNER_DIED | |
957 | bit and is then responsible for cleaning up the stale state left over by | |
1d09c150 | 958 | the dead owner. |
30239c10 MK |
959 | .\" tglx (July 2015): |
960 | .\" The FUTEX_OWNER_DIED bit can also be set on uncontended futexes, where | |
961 | .\" the kernel has no state associated. This happens via the robust futex | |
962 | .\" mechanism. In that case the futex value will be set to | |
963 | .\" FUTEX_OWNER_DIED. The robust futex mechanism is also available for non | |
964 | .\" PI futexes. | |
efeece04 | 965 | .PP |
601399f3 MK |
966 | PI futexes are operated on by specifying one of the values listed below in |
967 | .IR futex_op . | |
968 | Note that the PI futex operations must be used as paired operations | |
969 | and are subject to some additional requirements: | |
970 | .IP * 3 | |
eeeee811 AC |
971 | .BR FUTEX_LOCK_PI , |
972 | .BR FUTEX_LOCK_PI2 , | |
601399f3 MK |
973 | and |
974 | .B FUTEX_TRYLOCK_PI | |
975 | pair with | |
d8012462 | 976 | .BR FUTEX_UNLOCK_PI . |
601399f3 MK |
977 | .B FUTEX_UNLOCK_PI |
978 | must be called only on a futex owned by the calling thread, | |
979 | as defined by the value policy, otherwise the error | |
980 | .B EPERM | |
981 | results. | |
982 | .IP * | |
983 | .B FUTEX_WAIT_REQUEUE_PI | |
984 | pairs with | |
985 | .BR FUTEX_CMP_REQUEUE_PI . | |
986 | This must be performed from a non-PI futex to a distinct PI futex | |
987 | (or the error | |
988 | .B EINVAL | |
989 | results). | |
990 | Additionally, | |
991 | .I val | |
992 | (the number of waiters to be woken) must be 1 | |
993 | (or the error | |
994 | .B EINVAL | |
995 | results). | |
11ac5b51 | 996 | .PP |
601399f3 | 997 | The PI futex operations are as follows: |
70b06b90 MK |
998 | .\" |
999 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
1000 | .\" | |
d67e21f5 MK |
1001 | .TP |
1002 | .BR FUTEX_LOCK_PI " (since Linux 2.6.18)" | |
1003 | .\" commit c87e2837be82df479a6bae9f155c43516d2feebc | |
bc54ed38 | 1004 | This operation is used after an attempt to acquire |
fd105614 | 1005 | the lock via an atomic user-mode instruction failed |
4b35dc5d | 1006 | because the futex word has a nonzero value\(emspecifically, |
8297383e | 1007 | because it contained the (PID-namespace-specific) TID of the lock owner. |
efeece04 | 1008 | .IP |
4b35dc5d | 1009 | The operation checks the value of the futex word at the address |
67833bec | 1010 | .IR uaddr . |
70b06b90 MK |
1011 | If the value is 0, then the kernel tries to atomically set |
1012 | the futex value to the caller's TID. | |
c3875d1d | 1013 | If the futex word's value is nonzero, |
67833bec | 1014 | the kernel atomically sets the |
e0547e70 | 1015 | .B FUTEX_WAITERS |
67833bec MK |
1016 | bit, which signals the futex owner that it cannot unlock the futex in |
1017 | user space atomically by setting the futex value to 0. | |
c3875d1d MK |
1018 | .\" tglx (July 2015): |
1019 | .\" The operation here is similar to the FUTEX_WAIT logic. When the user | |
1020 | .\" space atomic acquire does not succeed because the futex value was non | |
1021 | .\" zero, then the waiter goes into the kernel, takes the kernel internal | |
1022 | .\" lock and retries the acquisition under the lock. If the acquisition | |
1023 | .\" does not succeed either, then it sets the FUTEX_WAITERS bit, to signal | |
1024 | .\" the lock owner that it needs to go into the kernel. Here is the pseudo | |
1025 | .\" code: | |
1026 | .\" | |
1027 | .\" lock(kernel_lock); | |
1028 | .\" retry: | |
9bfc9cb1 | 1029 | .\" |
c3875d1d | 1030 | .\" /* |
ba2c4752 | 1031 | .\" * Owner might have unlocked in user space before we |
c3875d1d MK |
1032 | .\" * were able to set the waiter bit. |
1033 | .\" */ | |
1034 | .\" if (atomic_acquire(futex) == SUCCESS) { | |
1035 | .\" unlock(kernel_lock()); | |
1036 | .\" return 0; | |
1037 | .\" } | |
1038 | .\" | |
1039 | .\" /* | |
1040 | .\" * Owner might have unlocked after the above atomic_acquire() | |
1041 | .\" * attempt. | |
1042 | .\" */ | |
1043 | .\" if (atomic_set_waiters_bit(futex) != SUCCESS) | |
1044 | .\" goto retry; | |
1045 | .\" | |
1046 | .\" queue_waiter(); | |
1047 | .\" unlock(kernel_lock); | |
1048 | .\" block(); | |
1049 | .\" | |
1050 | After that, the kernel: | |
1051 | .RS | |
1052 | .IP 1. 3 | |
1053 | Tries to find the thread which is associated with the owner TID. | |
1054 | .IP 2. | |
1055 | Creates or reuses kernel state on behalf of the owner. | |
1056 | (If this is the first waiter, there is no kernel state for this | |
1057 | futex, so kernel state is created by locking the RT-mutex | |
1058 | and the futex owner is made the owner of the RT-mutex. | |
1059 | If there are existing waiters, then the existing state is reused.) | |
1060 | .IP 3. | |
ca4e5b2b | 1061 | Attaches the waiter to the futex |
c3875d1d MK |
1062 | (i.e., the waiter is enqueued on the RT-mutex waiter list). |
1063 | .RE | |
1064 | .IP | |
ac894879 MK |
1065 | If more than one waiter exists, |
1066 | the enqueueing of the waiter is in descending priority order. | |
1067 | (For information on priority ordering, see the discussion of the | |
1068 | .BR SCHED_DEADLINE , | |
1069 | .BR SCHED_FIFO , | |
1070 | and | |
1071 | .BR SCHED_RR | |
1072 | scheduling policies in | |
1073 | .BR sched (7).) | |
1074 | The owner inherits either the waiter's CPU bandwidth | |
1075 | (if the waiter is scheduled under the | |
1076 | .BR SCHED_DEADLINE | |
1077 | policy) or the waiter's priority (if the waiter is scheduled under the | |
1078 | .BR SCHED_RR | |
1079 | or | |
1080 | .BR SCHED_FIFO | |
1081 | policy). | |
1d09c150 MK |
1082 | .\" August 2015: |
1083 | .\" mtk: If the realm is restricted purely to SCHED_OTHER (SCHED_NORMAL) | |
1084 | .\" processes, does the nice value come into play also? | |
1085 | .\" | |
1086 | .\" tglx: No. SCHED_OTHER/NORMAL tasks are handled in FIFO order | |
c3875d1d | 1087 | This inheritance follows the lock chain in the case of nested locking |
ca4e5b2b MK |
1088 | .\" (i.e., task 1 blocks on lock A, held by task 2, |
1089 | .\" while task 2 blocks on lock B, held by task 3) | |
c3875d1d | 1090 | and performs deadlock detection. |
efeece04 | 1091 | .IP |
e0547e70 TG |
1092 | The |
1093 | .I timeout | |
9ce19cf1 | 1094 | argument provides a timeout for the lock attempt. |
8064bfa5 MK |
1095 | If |
1096 | .I timeout | |
1097 | is not NULL, the structure it points to specifies | |
1098 | an absolute timeout, measured against the | |
9ce19cf1 MK |
1099 | .BR CLOCK_REALTIME |
1100 | clock. | |
c082f385 MK |
1101 | .\" 2016-07-07 response from Thomas Gleixner on LKML: |
1102 | .\" From: Thomas Gleixner <tglx@linutronix.de> | |
1103 | .\" Date: 6 July 2016 at 20:57 | |
1104 | .\" Subject: Re: futex: Allow FUTEX_CLOCK_REALTIME with FUTEX_WAIT op | |
2ae96e8a | 1105 | .\" |
c082f385 MK |
1106 | .\" On Thu, 23 Jun 2016, Michael Kerrisk (man-pages) wrote: |
1107 | .\" > On 06/23/2016 08:28 PM, Darren Hart wrote: | |
1108 | .\" > > And as a follow-on, what is the reason for FUTEX_LOCK_PI only using | |
1109 | .\" > > CLOCK_REALTIME? It seems reasonable to me that a user may want to wait a | |
1110 | .\" > > specific amount of time, regardless of wall time. | |
1111 | .\" > | |
1112 | .\" > Yes, that's another weird inconsistency. | |
2ae96e8a | 1113 | .\" |
c082f385 MK |
1114 | .\" The reason is that phtread_mutex_timedlock() uses absolute timeouts based on |
1115 | .\" CLOCK_REALTIME. glibc folks asked to make that the default behaviour back | |
1116 | .\" then when we added LOCK_PI. | |
9ce19cf1 MK |
1117 | If |
1118 | .I timeout | |
1119 | is NULL, the operation will block indefinitely. | |
efeece04 | 1120 | .IP |
a449c634 | 1121 | The |
e0547e70 TG |
1122 | .IR uaddr2 , |
1123 | .IR val , | |
1124 | and | |
eeeee811 | 1125 | .I val3 |
a449c634 | 1126 | arguments are ignored. |
67833bec | 1127 | .\" |
70b06b90 MK |
1128 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" |
1129 | .\" | |
d67e21f5 | 1130 | .TP |
eeeee811 | 1131 | .BR FUTEX_LOCK_PI2 " (since Linux 5.14)" |
e79977ae | 1132 | .\" commit bf22a6976897977b0a3f1aeba6823c959fc4fdae |
69bc3836 MK |
1133 | This operation is the same as |
1134 | .BR FUTEX_LOCK_PI , | |
1135 | except that the clock against which | |
eeeee811 | 1136 | .I timeout |
69bc3836 MK |
1137 | is measured is selectable. |
1138 | By default, the (absolute) timeout specified in | |
e79977ae | 1139 | .I timeout |
69bc3836 MK |
1140 | is measured againt the |
1141 | .B CLOCK_MONOTONIC | |
1142 | clock, but if the | |
1143 | .B FUTEX_CLOCK_REALTIME | |
1144 | flag is specified in | |
1145 | .IR futex_op , | |
1146 | then the timeout is measured against the | |
1147 | .B CLOCK_REALTIME | |
1148 | clock. | |
e79977ae KK |
1149 | .\" |
1150 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
1151 | .\" | |
1152 | .TP | |
12fdbe23 | 1153 | .BR FUTEX_TRYLOCK_PI " (since Linux 2.6.18)" |
d67e21f5 | 1154 | .\" commit c87e2837be82df479a6bae9f155c43516d2feebc |
3fbb1be1 | 1155 | This operation tries to acquire the lock at |
12fdbe23 | 1156 | .IR uaddr . |
c3875d1d MK |
1157 | It is invoked when a user-space atomic acquire did not |
1158 | succeed because the futex word was not 0. | |
efeece04 | 1159 | .IP |
8adaf0a7 MK |
1160 | Because the kernel has access to more state information than user space, |
1161 | acquisition of the lock might succeed if performed by the | |
1162 | kernel in cases where the futex word | |
1163 | (i.e., the state information accessible to use-space) contains stale state | |
c3875d1d MK |
1164 | .RB ( FUTEX_WAITERS |
1165 | and/or | |
1166 | .BR FUTEX_OWNER_DIED ). | |
1167 | This can happen when the owner of the futex died. | |
1d09c150 MK |
1168 | User space cannot handle this condition in a race-free manner, |
1169 | but the kernel can fix this up and acquire the futex. | |
ee65b0e8 MK |
1170 | .\" Paraphrasing a f2f conversation with Thomas Gleixner about the |
1171 | .\" above point (Aug 2015): ### | |
1172 | .\" There is a rare possibility of a race condition involving an | |
1173 | .\" uncontended futex with no owner, but with waiters. The | |
1174 | .\" kernel-user-space contract is that if a futex is nonzero, you must | |
1175 | .\" go into kernel. The futex was owned by a task, and that task dies | |
1176 | .\" but there are no waiters, so the futex value is non zero. | |
1177 | .\" Therefore, the next locker has to go into the kernel, | |
1178 | .\" so that the kernel has a chance to clean up. (CMXCH on zero | |
1179 | .\" in user space would fail, so kernel has to clean up.) | |
8adaf0a7 MK |
1180 | .\" Darren Hart (Oct 2015): |
1181 | .\" The trylock in the kernel has more state, so it can independently | |
ba2c4752 | 1182 | .\" verify the flags that user space must trust implicitly. |
efeece04 | 1183 | .IP |
084744ef MK |
1184 | The |
1185 | .IR uaddr2 , | |
1186 | .IR val , | |
1187 | .IR timeout , | |
1188 | and | |
1189 | .IR val3 | |
1190 | arguments are ignored. | |
70b06b90 MK |
1191 | .\" |
1192 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
1193 | .\" | |
d67e21f5 | 1194 | .TP |
12fdbe23 | 1195 | .BR FUTEX_UNLOCK_PI " (since Linux 2.6.18)" |
d67e21f5 | 1196 | .\" commit c87e2837be82df479a6bae9f155c43516d2feebc |
d4ba4328 | 1197 | This operation wakes the top priority waiter that is waiting in |
ecae2099 | 1198 | .B FUTEX_LOCK_PI |
e79977ae KK |
1199 | or |
1200 | .B FUTEX_LOCK_PI2 | |
ecae2099 TG |
1201 | on the futex address provided by the |
1202 | .I uaddr | |
1203 | argument. | |
efeece04 | 1204 | .IP |
1d09c150 | 1205 | This is called when the user-space value at |
ecae2099 TG |
1206 | .I uaddr |
1207 | cannot be changed atomically from a TID (of the owner) to 0. | |
efeece04 | 1208 | .IP |
ecae2099 TG |
1209 | The |
1210 | .IR uaddr2 , | |
1211 | .IR val , | |
1212 | .IR timeout , | |
1213 | and | |
1214 | .IR val3 | |
11a194bf | 1215 | arguments are ignored. |
70b06b90 MK |
1216 | .\" |
1217 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
1218 | .\" | |
d67e21f5 | 1219 | .TP |
d67e21f5 MK |
1220 | .BR FUTEX_CMP_REQUEUE_PI " (since Linux 2.6.31)" |
1221 | .\" commit 52400ba946759af28442dee6265c5c0180ac7122 | |
f812a08b DH |
1222 | This operation is a PI-aware variant of |
1223 | .BR FUTEX_CMP_REQUEUE . | |
1224 | It requeues waiters that are blocked via | |
1225 | .B FUTEX_WAIT_REQUEUE_PI | |
1226 | on | |
1227 | .I uaddr | |
1228 | from a non-PI source futex | |
1229 | .RI ( uaddr ) | |
1230 | to a PI target futex | |
1231 | .RI ( uaddr2 ). | |
efeece04 | 1232 | .IP |
9e54d26d MK |
1233 | As with |
1234 | .BR FUTEX_CMP_REQUEUE , | |
1235 | this operation wakes up a maximum of | |
1236 | .I val | |
1237 | waiters that are waiting on the futex at | |
1238 | .IR uaddr . | |
1239 | However, for | |
1240 | .BR FUTEX_CMP_REQUEUE_PI , | |
1241 | .I val | |
6fbeb8f4 | 1242 | is required to be 1 |
939ca89f | 1243 | (since the main point is to avoid a thundering herd). |
9e54d26d MK |
1244 | The remaining waiters are removed from the wait queue of the source futex at |
1245 | .I uaddr | |
1246 | and added to the wait queue of the target futex at | |
1247 | .IR uaddr2 . | |
efeece04 | 1248 | .IP |
9e54d26d | 1249 | The |
768d3c23 | 1250 | .I val2 |
c6d8cf21 MK |
1251 | .\" val2 is the cap on the number of requeued waiters. |
1252 | .\" In the glibc pthread_cond_broadcast() implementation, this argument | |
1253 | .\" is specified as INT_MAX, and for pthread_cond_signal() it is 0. | |
9e54d26d | 1254 | and |
768d3c23 | 1255 | .I val3 |
9e54d26d MK |
1256 | arguments serve the same purposes as for |
1257 | .BR FUTEX_CMP_REQUEUE . | |
70b06b90 | 1258 | .\" |
8297383e | 1259 | .\" The page at http://locklessinc.com/articles/futex_cheat_sheet/ |
be376673 | 1260 | .\" notes that "priority-inheritance Futex to priority-inheritance |
8297383e MK |
1261 | .\" Futex requeues are currently unsupported". However, probably |
1262 | .\" the page does not need to say nothing about this, since | |
1263 | .\" Thomas Gleixner commented (July 2015): "they never will be | |
1264 | .\" supported because they make no sense at all" | |
70b06b90 MK |
1265 | .\" |
1266 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
1267 | .\" | |
d67e21f5 MK |
1268 | .TP |
1269 | .BR FUTEX_WAIT_REQUEUE_PI " (since Linux 2.6.31)" | |
1270 | .\" commit 52400ba946759af28442dee6265c5c0180ac7122 | |
70b06b90 | 1271 | .\" |
c3875d1d | 1272 | Wait on a non-PI futex at |
6ff1b4c0 | 1273 | .I uaddr |
c3875d1d MK |
1274 | and potentially be requeued (via a |
1275 | .BR FUTEX_CMP_REQUEUE_PI | |
1276 | operation in another task) onto a PI futex at | |
6ff1b4c0 TG |
1277 | .IR uaddr2 . |
1278 | The wait operation on | |
1279 | .I uaddr | |
c3875d1d | 1280 | is the same as for |
6ff1b4c0 | 1281 | .BR FUTEX_WAIT . |
efeece04 | 1282 | .IP |
6ff1b4c0 TG |
1283 | The waiter can be removed from the wait on |
1284 | .I uaddr | |
6ff1b4c0 | 1285 | without requeueing on |
c3875d1d MK |
1286 | .IR uaddr2 |
1287 | via a | |
1d09c150 | 1288 | .BR FUTEX_WAKE |
c3875d1d MK |
1289 | operation in another task. |
1290 | In this case, the | |
1291 | .BR FUTEX_WAIT_REQUEUE_PI | |
3fbb1be1 MK |
1292 | operation fails with the error |
1293 | .BR EAGAIN . | |
efeece04 | 1294 | .IP |
63bea7dc MK |
1295 | If |
1296 | .I timeout | |
8064bfa5 MK |
1297 | is not NULL, the structure it points to specifies |
1298 | an absolute timeout for the wait operation. | |
63bea7dc MK |
1299 | If |
1300 | .I timeout | |
1301 | is NULL, the operation can block indefinitely. | |
efeece04 | 1302 | .IP |
a4e69912 MK |
1303 | The |
1304 | .I val3 | |
1305 | argument is ignored. | |
efeece04 | 1306 | .IP |
abb571e8 MK |
1307 | The |
1308 | .BR FUTEX_WAIT_REQUEUE_PI | |
1309 | and | |
1310 | .BR FUTEX_CMP_REQUEUE_PI | |
1311 | were added to support a fairly specific use case: | |
1312 | support for priority-inheritance-aware POSIX threads condition variables. | |
1313 | The idea is that these operations should always be paired, | |
1314 | in order to ensure that user space and the kernel remain in sync. | |
1315 | Thus, in the | |
1316 | .BR FUTEX_WAIT_REQUEUE_PI | |
1317 | operation, the user-space application pre-specifies the target | |
1318 | of the requeue that takes place in the | |
1319 | .BR FUTEX_CMP_REQUEUE_PI | |
1320 | operation. | |
1321 | .\" | |
1322 | .\" Darren Hart notes that a patch to allow glibc to fully support | |
1af427a4 | 1323 | .\" PI-aware pthreads condition variables has not yet been accepted into |
abb571e8 MK |
1324 | .\" glibc. The story is complex, and can be found at |
1325 | .\" https://sourceware.org/bugzilla/show_bug.cgi?id=11588 | |
1326 | .\" Darren notes that in the meantime, the patch is shipped with various | |
1af427a4 | 1327 | .\" PREEMPT_RT-enabled Linux systems. |
abb571e8 MK |
1328 | .\" |
1329 | .\" Related to the preceding, Darren proposed that somewhere, man-pages | |
1330 | .\" should document the following point: | |
1af427a4 | 1331 | .\" |
4c8cb0ff MK |
1332 | .\" While the Linux kernel, since 2.6.31, supports requeueing of |
1333 | .\" priority-inheritance (PI) aware mutexes via the | |
1334 | .\" FUTEX_WAIT_REQUEUE_PI and FUTEX_CMP_REQUEUE_PI futex operations, | |
1335 | .\" the glibc implementation does not yet take full advantage of this. | |
1336 | .\" Specifically, the condvar internal data lock remains a non-PI aware | |
1337 | .\" mutex, regardless of the type of the pthread_mutex associated with | |
1338 | .\" the condvar. This can lead to an unbounded priority inversion on | |
1339 | .\" the internal data lock even when associating a PI aware | |
1340 | .\" pthread_mutex with a condvar during a pthread_cond*_wait | |
1341 | .\" operation. For this reason, it is not recommended to rely on | |
1342 | .\" priority inheritance when using pthread condition variables. | |
1af427a4 MK |
1343 | .\" |
1344 | .\" The problem is that the obvious location for this text is | |
1345 | .\" the pthread_cond*wait(3) man page. However, such a man page | |
abb571e8 | 1346 | .\" does not currently exist. |
70b06b90 | 1347 | .\" |
6700de24 | 1348 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" |
70b06b90 | 1349 | .\" |
47297adb | 1350 | .SH RETURN VALUE |
a5c5a06a MK |
1351 | In the event of an error (and assuming that |
1352 | .BR futex () | |
1353 | was invoked via | |
1354 | .BR syscall (2)), | |
1355 | all operations return \-1 and set | |
e808bba0 | 1356 | .I errno |
855d489a | 1357 | to indicate the error. |
efeece04 | 1358 | .PP |
e808bba0 MK |
1359 | The return value on success depends on the operation, |
1360 | as described in the following list: | |
fea681da MK |
1361 | .TP |
1362 | .B FUTEX_WAIT | |
077981d4 | 1363 | Returns 0 if the caller was woken up. |
4c8cb0ff MK |
1364 | Note that a wake-up can also be caused by common futex usage patterns |
1365 | in unrelated code that happened to have previously used the futex word's | |
1366 | memory location (e.g., typical futex-based implementations of | |
1367 | Pthreads mutexes can cause this under some conditions). | |
1368 | Therefore, callers should always conservatively assume that a return | |
1369 | value of 0 can mean a spurious wake-up, and use the futex word's value | |
bc54ed38 MK |
1370 | (i.e., the user-space synchronization scheme) |
1371 | to decide whether to continue to block or not. | |
fea681da MK |
1372 | .TP |
1373 | .B FUTEX_WAKE | |
bdc5957a | 1374 | Returns the number of waiters that were woken up. |
fea681da MK |
1375 | .TP |
1376 | .B FUTEX_FD | |
1377 | Returns the new file descriptor associated with the futex. | |
1378 | .TP | |
1379 | .B FUTEX_REQUEUE | |
bdc5957a | 1380 | Returns the number of waiters that were woken up. |
fea681da MK |
1381 | .TP |
1382 | .B FUTEX_CMP_REQUEUE | |
bdc5957a | 1383 | Returns the total number of waiters that were woken up or |
4b35dc5d | 1384 | requeued to the futex for the futex word at |
3dfcc11d MK |
1385 | .IR uaddr2 . |
1386 | If this value is greater than | |
1387 | .IR val , | |
fd105614 | 1388 | then the difference is the number of waiters requeued to the futex for the |
4c8cb0ff | 1389 | futex word at |
3dfcc11d | 1390 | .IR uaddr2 . |
dcad19c0 MK |
1391 | .TP |
1392 | .B FUTEX_WAKE_OP | |
a8b5b324 | 1393 | Returns the total number of waiters that were woken up. |
4c8cb0ff MK |
1394 | This is the sum of the woken waiters on the two futexes for |
1395 | the futex words at | |
a8b5b324 MK |
1396 | .I uaddr |
1397 | and | |
1398 | .IR uaddr2 . | |
dcad19c0 MK |
1399 | .TP |
1400 | .B FUTEX_WAIT_BITSET | |
077981d4 MK |
1401 | Returns 0 if the caller was woken up. |
1402 | See | |
4b35dc5d TR |
1403 | .B FUTEX_WAIT |
1404 | for how to interpret this correctly in practice. | |
dcad19c0 MK |
1405 | .TP |
1406 | .B FUTEX_WAKE_BITSET | |
bdc5957a | 1407 | Returns the number of waiters that were woken up. |
dcad19c0 MK |
1408 | .TP |
1409 | .B FUTEX_LOCK_PI | |
bf02a260 | 1410 | Returns 0 if the futex was successfully locked. |
dcad19c0 | 1411 | .TP |
e79977ae KK |
1412 | .B FUTEX_LOCK_PI2 |
1413 | Returns 0 if the futex was successfully locked. | |
1414 | .TP | |
dcad19c0 | 1415 | .B FUTEX_TRYLOCK_PI |
5c716eef | 1416 | Returns 0 if the futex was successfully locked. |
dcad19c0 MK |
1417 | .TP |
1418 | .B FUTEX_UNLOCK_PI | |
52bb928f | 1419 | Returns 0 if the futex was successfully unlocked. |
dcad19c0 MK |
1420 | .TP |
1421 | .B FUTEX_CMP_REQUEUE_PI | |
bdc5957a | 1422 | Returns the total number of waiters that were woken up or |
4b35dc5d | 1423 | requeued to the futex for the futex word at |
dddd395a MK |
1424 | .IR uaddr2 . |
1425 | If this value is greater than | |
1426 | .IR val , | |
4c8cb0ff MK |
1427 | then difference is the number of waiters requeued to the futex for |
1428 | the futex word at | |
dddd395a | 1429 | .IR uaddr2 . |
dcad19c0 MK |
1430 | .TP |
1431 | .B FUTEX_WAIT_REQUEUE_PI | |
4c8cb0ff MK |
1432 | Returns 0 if the caller was successfully requeued to the futex for |
1433 | the futex word at | |
22c15de9 | 1434 | .IR uaddr2 . |
70b06b90 MK |
1435 | .\" |
1436 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
1437 | .\" | |
fea681da MK |
1438 | .SH ERRORS |
1439 | .TP | |
1440 | .B EACCES | |
4b35dc5d | 1441 | No read access to the memory of a futex word. |
fea681da MK |
1442 | .TP |
1443 | .B EAGAIN | |
f48516d1 | 1444 | .RB ( FUTEX_WAIT , |
4b35dc5d | 1445 | .BR FUTEX_WAIT_BITSET , |
f48516d1 | 1446 | .BR FUTEX_WAIT_REQUEUE_PI ) |
badbf70c MK |
1447 | The value pointed to by |
1448 | .I uaddr | |
1449 | was not equal to the expected value | |
1450 | .I val | |
1451 | at the time of the call. | |
efeece04 | 1452 | .IP |
9732dd8b MK |
1453 | .BR Note : |
1454 | on Linux, the symbolic names | |
1455 | .B EAGAIN | |
1456 | and | |
1457 | .B EWOULDBLOCK | |
77da5feb | 1458 | (both of which appear in different parts of the kernel futex code) |
9732dd8b | 1459 | have the same value. |
badbf70c MK |
1460 | .TP |
1461 | .B EAGAIN | |
8f2068bb MK |
1462 | .RB ( FUTEX_CMP_REQUEUE , |
1463 | .BR FUTEX_CMP_REQUEUE_PI ) | |
ce5602fd | 1464 | The value pointed to by |
9f6c40c0 МК |
1465 | .I uaddr |
1466 | is not equal to the expected value | |
1467 | .IR val3 . | |
fea681da | 1468 | .TP |
eeeee811 | 1469 | .B EAGAIN |
5662f56a | 1470 | .RB ( FUTEX_LOCK_PI , |
e79977ae | 1471 | .BR FUTEX_LOCK_PI2 , |
aaec9032 MK |
1472 | .BR FUTEX_TRYLOCK_PI , |
1473 | .BR FUTEX_CMP_REQUEUE_PI ) | |
1474 | The futex owner thread ID of | |
1475 | .I uaddr | |
1476 | (for | |
1477 | .BR FUTEX_CMP_REQUEUE_PI : | |
1478 | .IR uaddr2 ) | |
1479 | is about to exit, | |
5662f56a MK |
1480 | but has not yet handled the internal state cleanup. |
1481 | Try again. | |
1482 | .TP | |
eeeee811 | 1483 | .B EDEADLK |
7a39e745 | 1484 | .RB ( FUTEX_LOCK_PI , |
e79977ae | 1485 | .BR FUTEX_LOCK_PI2 , |
9732dd8b MK |
1486 | .BR FUTEX_TRYLOCK_PI , |
1487 | .BR FUTEX_CMP_REQUEUE_PI ) | |
4b35dc5d | 1488 | The futex word at |
7a39e745 MK |
1489 | .I uaddr |
1490 | is already locked by the caller. | |
1491 | .TP | |
662c0da8 | 1492 | .BR EDEADLK |
c3875d1d | 1493 | .\" FIXME . I see that kernel/locking/rtmutex.c uses EDEADLK in some |
d6bb5a38 | 1494 | .\" places, and EDEADLOCK in others. On almost all architectures |
4c8cb0ff MK |
1495 | .\" these constants are synonymous. Is there a reason that both |
1496 | .\" names are used? | |
8297383e MK |
1497 | .\" |
1498 | .\" tglx (July 2015): "No. We should probably fix that." | |
1499 | .\" | |
662c0da8 | 1500 | .RB ( FUTEX_CMP_REQUEUE_PI ) |
4b35dc5d | 1501 | While requeueing a waiter to the PI futex for the futex word at |
662c0da8 MK |
1502 | .IR uaddr2 , |
1503 | the kernel detected a deadlock. | |
1504 | .TP | |
fea681da | 1505 | .B EFAULT |
1ea901e8 MK |
1506 | A required pointer argument (i.e., |
1507 | .IR uaddr , | |
1508 | .IR uaddr2 , | |
1509 | or | |
1510 | .IR timeout ) | |
496df304 | 1511 | did not point to a valid user-space address. |
fea681da | 1512 | .TP |
9f6c40c0 | 1513 | .B EINTR |
e808bba0 | 1514 | A |
9f6c40c0 | 1515 | .B FUTEX_WAIT |
2674f781 MK |
1516 | or |
1517 | .B FUTEX_WAIT_BITSET | |
e808bba0 | 1518 | operation was interrupted by a signal (see |
f529fd20 MK |
1519 | .BR signal (7)). |
1520 | In kernels before Linux 2.6.22, this error could also be returned for | |
b5fff4ea | 1521 | a spurious wakeup; since Linux 2.6.22, this no longer happens. |
9f6c40c0 | 1522 | .TP |
fea681da | 1523 | .B EINVAL |
180f97b7 | 1524 | The operation in |
eeeee811 | 1525 | .I futex_op |
180f97b7 | 1526 | is one of those that employs a timeout, but the supplied |
fb2f4c27 MK |
1527 | .I timeout |
1528 | argument was invalid | |
1529 | .RI ( tv_sec | |
1530 | was less than zero, or | |
eeeee811 | 1531 | .I tv_nsec |
cabee29d | 1532 | was not less than 1,000,000,000). |
fb2f4c27 MK |
1533 | .TP |
1534 | .B EINVAL | |
0c74df0b | 1535 | The operation specified in |
eeeee811 | 1536 | .I futex_op |
0c74df0b | 1537 | employs one or both of the pointers |
51ee94be | 1538 | .I uaddr |
a1f47699 | 1539 | and |
0c74df0b MK |
1540 | .IR uaddr2 , |
1541 | but one of these does not point to a valid object\(emthat is, | |
1542 | the address is not four-byte-aligned. | |
51ee94be MK |
1543 | .TP |
1544 | .B EINVAL | |
55cc422d TG |
1545 | .RB ( FUTEX_WAIT_BITSET , |
1546 | .BR FUTEX_WAKE_BITSET ) | |
5e1456d4 | 1547 | The bit mask supplied in |
eeeee811 | 1548 | .I val3 |
79c9b436 TG |
1549 | is zero. |
1550 | .TP | |
1551 | .B EINVAL | |
2abcba67 | 1552 | .RB ( FUTEX_CMP_REQUEUE_PI ) |
add875c0 MK |
1553 | .I uaddr |
1554 | equals | |
eeeee811 | 1555 | .I uaddr2 |
add875c0 MK |
1556 | (i.e., an attempt was made to requeue to the same futex). |
1557 | .TP | |
eeeee811 | 1558 | .B EINVAL |
ff597681 MK |
1559 | .RB ( FUTEX_FD ) |
1560 | The signal number supplied in | |
1561 | .I val | |
1562 | is invalid. | |
1563 | .TP | |
6bac3b85 | 1564 | .B EINVAL |
476debd7 MK |
1565 | .RB ( FUTEX_WAKE , |
1566 | .BR FUTEX_WAKE_OP , | |
1567 | .BR FUTEX_WAKE_BITSET , | |
1568 | .BR FUTEX_REQUEUE , | |
1569 | .BR FUTEX_CMP_REQUEUE ) | |
1570 | The kernel detected an inconsistency between the user-space state at | |
1571 | .I uaddr | |
1572 | and the kernel state\(emthat is, it detected a waiter which waits in | |
eeeee811 | 1573 | .B FUTEX_LOCK_PI |
e79977ae | 1574 | or |
eeeee811 | 1575 | .B FUTEX_LOCK_PI2 |
476debd7 MK |
1576 | on |
1577 | .IR uaddr . | |
1578 | .TP | |
1579 | .B EINVAL | |
a218ef20 | 1580 | .RB ( FUTEX_LOCK_PI , |
e79977ae | 1581 | .BR FUTEX_LOCK_PI2 , |
ce022f18 MK |
1582 | .BR FUTEX_TRYLOCK_PI , |
1583 | .BR FUTEX_UNLOCK_PI ) | |
a218ef20 MK |
1584 | The kernel detected an inconsistency between the user-space state at |
1585 | .I uaddr | |
1586 | and the kernel state. | |
ce022f18 | 1587 | This indicates either state corruption |
ce022f18 | 1588 | or that the kernel found a waiter on |
a218ef20 MK |
1589 | .I uaddr |
1590 | which is waiting via | |
eeeee811 | 1591 | .B FUTEX_WAIT |
a218ef20 MK |
1592 | or |
1593 | .BR FUTEX_WAIT_BITSET . | |
1594 | .TP | |
1595 | .B EINVAL | |
f9250b1a MK |
1596 | .RB ( FUTEX_CMP_REQUEUE_PI ) |
1597 | The kernel detected an inconsistency between the user-space state at | |
99c0041d MK |
1598 | .I uaddr2 |
1599 | and the kernel state; | |
ee65b0e8 MK |
1600 | .\" From a conversation with Thomas Gleixner (Aug 2015): ### |
1601 | .\" The kernel sees: I have non PI state for a futex you tried to | |
1602 | .\" tell me was PI | |
99c0041d | 1603 | that is, the kernel detected a waiter which waits via |
eeeee811 | 1604 | .B FUTEX_WAIT |
8297383e | 1605 | or |
eeeee811 | 1606 | .B FUTEX_WAIT_BITSET |
99c0041d MK |
1607 | on |
1608 | .IR uaddr2 . | |
1609 | .TP | |
1610 | .B EINVAL | |
1611 | .RB ( FUTEX_CMP_REQUEUE_PI ) | |
1612 | The kernel detected an inconsistency between the user-space state at | |
f9250b1a MK |
1613 | .I uaddr |
1614 | and the kernel state; | |
1615 | that is, the kernel detected a waiter which waits via | |
eeeee811 | 1616 | .B FUTEX_WAIT |
99c0041d | 1617 | or |
eeeee811 | 1618 | .B FUTEX_WAIT_BITSET |
f9250b1a MK |
1619 | on |
1620 | .IR uaddr . | |
1621 | .TP | |
1622 | .B EINVAL | |
99c0041d | 1623 | .RB ( FUTEX_CMP_REQUEUE_PI ) |
75299c8d MK |
1624 | The kernel detected an inconsistency between the user-space state at |
1625 | .I uaddr | |
1626 | and the kernel state; | |
1627 | that is, the kernel detected a waiter which waits on | |
1628 | .I uaddr | |
1629 | via | |
eeeee811 | 1630 | .B FUTEX_LOCK_PI |
e79977ae | 1631 | or |
eeeee811 | 1632 | .B FUTEX_LOCK_PI2 |
75299c8d MK |
1633 | (instead of |
1634 | .BR FUTEX_WAIT_REQUEUE_PI ). | |
99c0041d MK |
1635 | .TP |
1636 | .B EINVAL | |
9786b3ca | 1637 | .RB ( FUTEX_CMP_REQUEUE_PI ) |
8297383e MK |
1638 | .\" This deals with the case: |
1639 | .\" wait_requeue_pi(A, B); | |
1640 | .\" requeue_pi(A, C); | |
9786b3ca MK |
1641 | An attempt was made to requeue a waiter to a futex other than that |
1642 | specified by the matching | |
1643 | .B FUTEX_WAIT_REQUEUE_PI | |
1644 | call for that waiter. | |
1645 | .TP | |
1646 | .B EINVAL | |
f0c0d61c MK |
1647 | .RB ( FUTEX_CMP_REQUEUE_PI ) |
1648 | The | |
1649 | .I val | |
1650 | argument is not 1. | |
1651 | .TP | |
1652 | .B EINVAL | |
4832b48a | 1653 | Invalid argument. |
fea681da | 1654 | .TP |
d07d4ef3 MK |
1655 | .B ENFILE |
1656 | .RB ( FUTEX_FD ) | |
1657 | The system-wide limit on the total number of open files has been reached. | |
1658 | .TP | |
eeeee811 | 1659 | .B ENOMEM |
a449c634 | 1660 | .RB ( FUTEX_LOCK_PI , |
e79977ae | 1661 | .BR FUTEX_LOCK_PI2 , |
e34a8fb6 MK |
1662 | .BR FUTEX_TRYLOCK_PI , |
1663 | .BR FUTEX_CMP_REQUEUE_PI ) | |
a449c634 MK |
1664 | The kernel could not allocate memory to hold state information. |
1665 | .TP | |
4701fc28 MK |
1666 | .B ENOSYS |
1667 | Invalid operation specified in | |
d33602c4 | 1668 | .IR futex_op . |
9f6c40c0 | 1669 | .TP |
4a7e5b05 MK |
1670 | .B ENOSYS |
1671 | The | |
eeeee811 | 1672 | .B FUTEX_CLOCK_REALTIME |
4a7e5b05 | 1673 | option was specified in |
1afcee7c | 1674 | .IR futex_op , |
4a7e5b05 | 1675 | but the accompanying operation was neither |
017d194b MK |
1676 | .BR FUTEX_WAIT , |
1677 | .BR FUTEX_WAIT_BITSET , | |
e79977ae | 1678 | .BR FUTEX_WAIT_REQUEUE_PI , |
4a7e5b05 | 1679 | nor |
e79977ae | 1680 | .BR FUTEX_LOCK_PI2 . |
4a7e5b05 | 1681 | .TP |
eeeee811 | 1682 | .B ENOSYS |
a9dcb4d1 | 1683 | .RB ( FUTEX_LOCK_PI , |
e79977ae | 1684 | .BR FUTEX_LOCK_PI2 , |
f2424fae | 1685 | .BR FUTEX_TRYLOCK_PI , |
4945ff19 | 1686 | .BR FUTEX_UNLOCK_PI , |
4cf92894 | 1687 | .BR FUTEX_CMP_REQUEUE_PI , |
794bb106 | 1688 | .BR FUTEX_WAIT_REQUEUE_PI ) |
4b35dc5d | 1689 | A run-time check determined that the operation is not available. |
f0a9e8f4 | 1690 | The PI-futex operations are not implemented on all architectures and |
077981d4 | 1691 | are not supported on some CPU variants. |
a9dcb4d1 | 1692 | .TP |
eeeee811 | 1693 | .B EPERM |
c7589177 | 1694 | .RB ( FUTEX_LOCK_PI , |
e79977ae | 1695 | .BR FUTEX_LOCK_PI2 , |
dc2742a8 MK |
1696 | .BR FUTEX_TRYLOCK_PI , |
1697 | .BR FUTEX_CMP_REQUEUE_PI ) | |
04331c3f | 1698 | The caller is not allowed to attach itself to the futex at |
dc2742a8 MK |
1699 | .I uaddr |
1700 | (for | |
1701 | .BR FUTEX_CMP_REQUEUE_PI : | |
1702 | the futex at | |
1703 | .IR uaddr2 ). | |
c7589177 MK |
1704 | (This may be caused by a state corruption in user space.) |
1705 | .TP | |
eeeee811 | 1706 | .B EPERM |
87276709 | 1707 | .RB ( FUTEX_UNLOCK_PI ) |
4b35dc5d | 1708 | The caller does not own the lock represented by the futex word. |
76f347ba | 1709 | .TP |
eeeee811 | 1710 | .B ESRCH |
0b0e4934 | 1711 | .RB ( FUTEX_LOCK_PI , |
e79977ae | 1712 | .BR FUTEX_LOCK_PI2 , |
9732dd8b MK |
1713 | .BR FUTEX_TRYLOCK_PI , |
1714 | .BR FUTEX_CMP_REQUEUE_PI ) | |
4b35dc5d | 1715 | The thread ID in the futex word at |
0b0e4934 MK |
1716 | .I uaddr |
1717 | does not exist. | |
1718 | .TP | |
eeeee811 | 1719 | .B ESRCH |
360f773c | 1720 | .RB ( FUTEX_CMP_REQUEUE_PI ) |
4b35dc5d | 1721 | The thread ID in the futex word at |
360f773c MK |
1722 | .I uaddr2 |
1723 | does not exist. | |
1724 | .TP | |
9f6c40c0 | 1725 | .B ETIMEDOUT |
4d85047f | 1726 | The operation in |
eeeee811 | 1727 | .I futex_op |
4d85047f MK |
1728 | employed the timeout specified in |
1729 | .IR timeout , | |
1730 | and the timeout expired before the operation completed. | |
70b06b90 MK |
1731 | .\" |
1732 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
1733 | .\" | |
47297adb | 1734 | .SH VERSIONS |
81c9d87e MK |
1735 | Futexes were first made available in a stable kernel release |
1736 | with Linux 2.6.0. | |
efeece04 | 1737 | .PP |
4c8cb0ff MK |
1738 | Initial futex support was merged in Linux 2.5.7 but with different |
1739 | semantics from what was described above. | |
52dee70e | 1740 | A four-argument system call with the semantics |
fd3fa7ef | 1741 | described in this page was introduced in Linux 2.5.40. |
d0442d14 MK |
1742 | A fifth argument was added in Linux 2.5.70, |
1743 | and a sixth argument was added in Linux 2.6.7. | |
47297adb | 1744 | .SH CONFORMING TO |
8382f16d | 1745 | This system call is Linux-specific. |
47297adb | 1746 | .SH NOTES |
02f7b623 | 1747 | Several higher-level programming abstractions are implemented via futexes, |
e24fbf10 | 1748 | including POSIX semaphores and |
02f7b623 MK |
1749 | various POSIX threads synchronization mechanisms |
1750 | (mutexes, condition variables, read-write locks, and barriers). | |
74f58a64 MK |
1751 | .\" TODO FIXME(Torvald) Above, we cite this section and claim it contains |
1752 | .\" details on the synchronization semantics; add the C11 equivalents | |
1753 | .\" here (or whatever we find consensus for). | |
305cc415 MK |
1754 | .\" |
1755 | .\"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" | |
1756 | .\" | |
a14af333 | 1757 | .SH EXAMPLES |
bc54ed38 MK |
1758 | The program below demonstrates use of futexes in a program where a parent |
1759 | process and a child process use a pair of futexes located inside a | |
305cc415 MK |
1760 | shared anonymous mapping to synchronize access to a shared resource: |
1761 | the terminal. | |
1762 | The two processes each write | |
1763 | .IR nloops | |
1764 | (a command-line argument that defaults to 5 if omitted) | |
1765 | messages to the terminal and employ a synchronization protocol | |
1766 | that ensures that they alternate in writing messages. | |
1767 | Upon running this program we see output such as the following: | |
efeece04 | 1768 | .PP |
305cc415 | 1769 | .in +4n |
b76974c1 | 1770 | .EX |
305cc415 MK |
1771 | $ \fB./futex_demo\fP |
1772 | Parent (18534) 0 | |
1773 | Child (18535) 0 | |
1774 | Parent (18534) 1 | |
1775 | Child (18535) 1 | |
1776 | Parent (18534) 2 | |
1777 | Child (18535) 2 | |
1778 | Parent (18534) 3 | |
1779 | Child (18535) 3 | |
1780 | Parent (18534) 4 | |
1781 | Child (18535) 4 | |
b76974c1 | 1782 | .EE |
305cc415 MK |
1783 | .in |
1784 | .SS Program source | |
1785 | \& | |
e7d0bb47 | 1786 | .EX |
305cc415 MK |
1787 | /* futex_demo.c |
1788 | ||
1789 | Usage: futex_demo [nloops] | |
1790 | (Default: 5) | |
1791 | ||
1792 | Demonstrate the use of futexes in a program where parent and child | |
1793 | use a pair of futexes located inside a shared anonymous mapping to | |
1794 | synchronize access to a shared resource: the terminal. The two | |
1795 | processes each write \(aqnum\-loops\(aq messages to the terminal and employ | |
1796 | a synchronization protocol that ensures that they alternate in | |
1797 | writing messages. | |
1798 | */ | |
1799 | #define _GNU_SOURCE | |
1800 | #include <stdio.h> | |
1801 | #include <errno.h> | |
915c4ba3 | 1802 | #include <stdatomic.h> |
8eb90116 | 1803 | #include <stdint.h> |
305cc415 MK |
1804 | #include <stdlib.h> |
1805 | #include <unistd.h> | |
1806 | #include <sys/wait.h> | |
1807 | #include <sys/mman.h> | |
1808 | #include <sys/syscall.h> | |
1809 | #include <linux/futex.h> | |
1810 | #include <sys/time.h> | |
1811 | ||
d1a71985 | 1812 | #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \e |
305cc415 MK |
1813 | } while (0) |
1814 | ||
2253ecf0 | 1815 | static uint32_t *futex1, *futex2, *iaddr; |
305cc415 MK |
1816 | |
1817 | static int | |
2253ecf0 AC |
1818 | futex(uint32_t *uaddr, int futex_op, uint32_t val, |
1819 | const struct timespec *timeout, uint32_t *uaddr2, uint32_t val3) | |
305cc415 MK |
1820 | { |
1821 | return syscall(SYS_futex, uaddr, futex_op, val, | |
c1e04f01 | 1822 | timeout, uaddr2, val3); |
305cc415 MK |
1823 | } |
1824 | ||
1825 | /* Acquire the futex pointed to by \(aqfutexp\(aq: wait for its value to | |
1826 | become 1, and then set the value to 0. */ | |
1827 | ||
1828 | static void | |
2253ecf0 | 1829 | fwait(uint32_t *futexp) |
305cc415 | 1830 | { |
2253ecf0 | 1831 | long s; |
305cc415 | 1832 | |
915c4ba3 BP |
1833 | /* atomic_compare_exchange_strong(ptr, oldval, newval) |
1834 | atomically performs the equivalent of: | |
305cc415 | 1835 | |
915c4ba3 | 1836 | if (*ptr == *oldval) |
305cc415 MK |
1837 | *ptr = newval; |
1838 | ||
915c4ba3 | 1839 | It returns true if the test yielded true and *ptr was updated. */ |
305cc415 | 1840 | |
305cc415 | 1841 | while (1) { |
83e80dda | 1842 | |
63ad44cb | 1843 | /* Is the futex available? */ |
2253ecf0 | 1844 | const uint32_t one = 1; |
09e456c2 | 1845 | if (atomic_compare_exchange_strong(futexp, &one, 0)) |
305cc415 MK |
1846 | break; /* Yes */ |
1847 | ||
c6beb8a1 | 1848 | /* Futex is not available; wait. */ |
83e80dda | 1849 | |
63ad44cb HS |
1850 | s = futex(futexp, FUTEX_WAIT, 0, NULL, NULL, 0); |
1851 | if (s == \-1 && errno != EAGAIN) | |
1852 | errExit("futex\-FUTEX_WAIT"); | |
305cc415 MK |
1853 | } |
1854 | } | |
1855 | ||
1856 | /* Release the futex pointed to by \(aqfutexp\(aq: if the futex currently | |
1857 | has the value 0, set its value to 1 and the wake any futex waiters, | |
e2c75104 | 1858 | so that if the peer is blocked in fwait(), it can proceed. */ |
305cc415 MK |
1859 | |
1860 | static void | |
2253ecf0 | 1861 | fpost(uint32_t *futexp) |
305cc415 | 1862 | { |
2253ecf0 | 1863 | long s; |
305cc415 | 1864 | |
68219aba | 1865 | /* atomic_compare_exchange_strong() was described |
c6beb8a1 | 1866 | in comments above. */ |
305cc415 | 1867 | |
2253ecf0 | 1868 | const uint32_t zero = 0; |
09e456c2 | 1869 | if (atomic_compare_exchange_strong(futexp, &zero, 1)) { |
305cc415 MK |
1870 | s = futex(futexp, FUTEX_WAKE, 1, NULL, NULL, 0); |
1871 | if (s == \-1) | |
1872 | errExit("futex\-FUTEX_WAKE"); | |
1873 | } | |
1874 | } | |
1875 | ||
1876 | int | |
1877 | main(int argc, char *argv[]) | |
1878 | { | |
1879 | pid_t childPid; | |
88893a77 | 1880 | int nloops; |
305cc415 MK |
1881 | |
1882 | setbuf(stdout, NULL); | |
1883 | ||
1884 | nloops = (argc > 1) ? atoi(argv[1]) : 5; | |
1885 | ||
1886 | /* Create a shared anonymous mapping that will hold the futexes. | |
1887 | Since the futexes are being shared between processes, we | |
1888 | subsequently use the "shared" futex operations (i.e., not the | |
c6beb8a1 | 1889 | ones suffixed "_PRIVATE"). */ |
305cc415 | 1890 | |
d60a7a9a | 1891 | iaddr = mmap(NULL, sizeof(*iaddr) * 2, PROT_READ | PROT_WRITE, |
305cc415 MK |
1892 | MAP_ANONYMOUS | MAP_SHARED, \-1, 0); |
1893 | if (iaddr == MAP_FAILED) | |
1894 | errExit("mmap"); | |
1895 | ||
1896 | futex1 = &iaddr[0]; | |
1897 | futex2 = &iaddr[1]; | |
1898 | ||
1899 | *futex1 = 0; /* State: unavailable */ | |
1900 | *futex2 = 1; /* State: available */ | |
1901 | ||
1902 | /* Create a child process that inherits the shared anonymous | |
c6beb8a1 | 1903 | mapping. */ |
305cc415 MK |
1904 | |
1905 | childPid = fork(); | |
92a46690 | 1906 | if (childPid == \-1) |
305cc415 MK |
1907 | errExit("fork"); |
1908 | ||
1909 | if (childPid == 0) { /* Child */ | |
88893a77 | 1910 | for (int j = 0; j < nloops; j++) { |
305cc415 | 1911 | fwait(futex1); |
8eb90116 | 1912 | printf("Child (%jd) %d\en", (intmax_t) getpid(), j); |
305cc415 MK |
1913 | fpost(futex2); |
1914 | } | |
1915 | ||
1916 | exit(EXIT_SUCCESS); | |
1917 | } | |
1918 | ||
c6beb8a1 | 1919 | /* Parent falls through to here. */ |
305cc415 | 1920 | |
88893a77 | 1921 | for (int j = 0; j < nloops; j++) { |
305cc415 | 1922 | fwait(futex2); |
8eb90116 | 1923 | printf("Parent (%jd) %d\en", (intmax_t) getpid(), j); |
305cc415 MK |
1924 | fpost(futex1); |
1925 | } | |
1926 | ||
1927 | wait(NULL); | |
1928 | ||
1929 | exit(EXIT_SUCCESS); | |
1930 | } | |
e7d0bb47 | 1931 | .EE |
47297adb | 1932 | .SH SEE ALSO |
4c222281 | 1933 | .ad l |
9913033c | 1934 | .BR get_robust_list (2), |
d806bc05 | 1935 | .BR restart_syscall (2), |
e0074751 | 1936 | .BR pthread_mutexattr_getprotocol (3), |
ac894879 MK |
1937 | .BR futex (7), |
1938 | .BR sched (7) | |
fea681da | 1939 | .PP |
f5ad572f MK |
1940 | The following kernel source files: |
1941 | .IP * 2 | |
b49c2acb | 1942 | .I Documentation/pi\-futex.txt |
f5ad572f | 1943 | .IP * |
b49c2acb | 1944 | .I Documentation/futex\-requeue\-pi.txt |
f5ad572f | 1945 | .IP * |
b49c2acb | 1946 | .I Documentation/locking/rt\-mutex.txt |
f5ad572f | 1947 | .IP * |
b49c2acb | 1948 | .I Documentation/locking/rt\-mutex\-design.txt |
8fe019c7 | 1949 | .IP * |
b49c2acb | 1950 | .I Documentation/robust\-futex\-ABI.txt |
43b99089 | 1951 | .PP |
4c222281 | 1952 | Franke, H., Russell, R., and Kirwood, M., 2002. |
52087dd3 | 1953 | \fIFuss, Futexes and Furwocks: Fast Userlevel Locking in Linux\fP |
4c222281 | 1954 | (from proceedings of the Ottawa Linux Symposium 2002), |
9b936e9e | 1955 | .br |
5465ae95 | 1956 | .UR http://kernel.org\:/doc\:/ols\:/2002\:/ols2002\-pages\-479\-495.pdf |
608bf950 | 1957 | .UE |
efeece04 | 1958 | .PP |
4c222281 | 1959 | Hart, D., 2009. \fIA futex overview and update\fP, |
2ed26199 MK |
1960 | .UR http://lwn.net/Articles/360699/ |
1961 | .UE | |
efeece04 | 1962 | .PP |
8fb01fde | 1963 | Hart, D.\& and Guniguntala, D., 2009. |
0483b6cc | 1964 | \fIRequeue-PI: Making Glibc Condvars PI-Aware\fP |
4c222281 | 1965 | (from proceedings of the 2009 Real-Time Linux Workshop), |
0483b6cc MK |
1966 | .UR http://lwn.net/images/conf/rtlws11/papers/proc/p10.pdf |
1967 | .UE | |
efeece04 | 1968 | .PP |
4c222281 | 1969 | Drepper, U., 2011. \fIFutexes Are Tricky\fP, |
f42eb21b MK |
1970 | .UR http://www.akkadia.org/drepper/futex.pdf |
1971 | .UE | |
9b936e9e MK |
1972 | .PP |
1973 | Futex example library, futex-*.tar.bz2 at | |
1974 | .br | |
a605264d | 1975 | .UR ftp://ftp.kernel.org\:/pub\:/linux\:/kernel\:/people\:/rusty/ |
608bf950 | 1976 | .UE |
34f14794 | 1977 | .\" |
74f58a64 | 1978 | .\" FIXME(Torvald) We should probably refer to the glibc code here, in |
9915ea23 MK |
1979 | .\" particular the glibc-internal futex wrapper functions that are |
1980 | .\" WIP, and the generic pthread_mutex_t and perhaps condvar | |
1981 | .\" implementations. |