Commit | Line | Data |
---|---|---|
108b42b4 DH |
1 | ============================ |
2 | LINUX KERNEL MEMORY BARRIERS | |
3 | ============================ | |
4 | ||
5 | By: David Howells <dhowells@redhat.com> | |
90fddabf | 6 | Paul E. McKenney <paulmck@linux.vnet.ibm.com> |
108b42b4 DH |
7 | |
8 | Contents: | |
9 | ||
10 | (*) Abstract memory access model. | |
11 | ||
12 | - Device operations. | |
13 | - Guarantees. | |
14 | ||
15 | (*) What are memory barriers? | |
16 | ||
17 | - Varieties of memory barrier. | |
18 | - What may not be assumed about memory barriers? | |
19 | - Data dependency barriers. | |
20 | - Control dependencies. | |
21 | - SMP barrier pairing. | |
22 | - Examples of memory barrier sequences. | |
670bd95e | 23 | - Read memory barriers vs load speculation. |
241e6663 | 24 | - Transitivity |
108b42b4 DH |
25 | |
26 | (*) Explicit kernel barriers. | |
27 | ||
28 | - Compiler barrier. | |
81fc6323 | 29 | - CPU memory barriers. |
108b42b4 DH |
30 | - MMIO write barrier. |
31 | ||
32 | (*) Implicit kernel memory barriers. | |
33 | ||
34 | - Locking functions. | |
35 | - Interrupt disabling functions. | |
50fa610a | 36 | - Sleep and wake-up functions. |
108b42b4 DH |
37 | - Miscellaneous functions. |
38 | ||
39 | (*) Inter-CPU locking barrier effects. | |
40 | ||
41 | - Locks vs memory accesses. | |
42 | - Locks vs I/O accesses. | |
43 | ||
44 | (*) Where are memory barriers needed? | |
45 | ||
46 | - Interprocessor interaction. | |
47 | - Atomic operations. | |
48 | - Accessing devices. | |
49 | - Interrupts. | |
50 | ||
51 | (*) Kernel I/O barrier effects. | |
52 | ||
53 | (*) Assumed minimum execution ordering model. | |
54 | ||
55 | (*) The effects of the cpu cache. | |
56 | ||
57 | - Cache coherency. | |
58 | - Cache coherency vs DMA. | |
59 | - Cache coherency vs MMIO. | |
60 | ||
61 | (*) The things CPUs get up to. | |
62 | ||
63 | - And then there's the Alpha. | |
64 | ||
90fddabf DH |
65 | (*) Example uses. |
66 | ||
67 | - Circular buffers. | |
68 | ||
108b42b4 DH |
69 | (*) References. |
70 | ||
71 | ||
72 | ============================ | |
73 | ABSTRACT MEMORY ACCESS MODEL | |
74 | ============================ | |
75 | ||
76 | Consider the following abstract model of the system: | |
77 | ||
78 | : : | |
79 | : : | |
80 | : : | |
81 | +-------+ : +--------+ : +-------+ | |
82 | | | : | | : | | | |
83 | | | : | | : | | | |
84 | | CPU 1 |<----->| Memory |<----->| CPU 2 | | |
85 | | | : | | : | | | |
86 | | | : | | : | | | |
87 | +-------+ : +--------+ : +-------+ | |
88 | ^ : ^ : ^ | |
89 | | : | : | | |
90 | | : | : | | |
91 | | : v : | | |
92 | | : +--------+ : | | |
93 | | : | | : | | |
94 | | : | | : | | |
95 | +---------->| Device |<----------+ | |
96 | : | | : | |
97 | : | | : | |
98 | : +--------+ : | |
99 | : : | |
100 | ||
101 | Each CPU executes a program that generates memory access operations. In the | |
102 | abstract CPU, memory operation ordering is very relaxed, and a CPU may actually | |
103 | perform the memory operations in any order it likes, provided program causality | |
104 | appears to be maintained. Similarly, the compiler may also arrange the | |
105 | instructions it emits in any order it likes, provided it doesn't affect the | |
106 | apparent operation of the program. | |
107 | ||
108 | So in the above diagram, the effects of the memory operations performed by a | |
109 | CPU are perceived by the rest of the system as the operations cross the | |
110 | interface between the CPU and rest of the system (the dotted lines). | |
111 | ||
112 | ||
113 | For example, consider the following sequence of events: | |
114 | ||
115 | CPU 1 CPU 2 | |
116 | =============== =============== | |
117 | { A == 1; B == 2 } | |
118 | A = 3; x = A; | |
119 | B = 4; y = B; | |
120 | ||
121 | The set of accesses as seen by the memory system in the middle can be arranged | |
122 | in 24 different combinations: | |
123 | ||
124 | STORE A=3, STORE B=4, x=LOAD A->3, y=LOAD B->4 | |
125 | STORE A=3, STORE B=4, y=LOAD B->4, x=LOAD A->3 | |
126 | STORE A=3, x=LOAD A->3, STORE B=4, y=LOAD B->4 | |
127 | STORE A=3, x=LOAD A->3, y=LOAD B->2, STORE B=4 | |
128 | STORE A=3, y=LOAD B->2, STORE B=4, x=LOAD A->3 | |
129 | STORE A=3, y=LOAD B->2, x=LOAD A->3, STORE B=4 | |
130 | STORE B=4, STORE A=3, x=LOAD A->3, y=LOAD B->4 | |
131 | STORE B=4, ... | |
132 | ... | |
133 | ||
134 | and can thus result in four different combinations of values: | |
135 | ||
136 | x == 1, y == 2 | |
137 | x == 1, y == 4 | |
138 | x == 3, y == 2 | |
139 | x == 3, y == 4 | |
140 | ||
141 | ||
142 | Furthermore, the stores committed by a CPU to the memory system may not be | |
143 | perceived by the loads made by another CPU in the same order as the stores were | |
144 | committed. | |
145 | ||
146 | ||
147 | As a further example, consider this sequence of events: | |
148 | ||
149 | CPU 1 CPU 2 | |
150 | =============== =============== | |
151 | { A == 1, B == 2, C = 3, P == &A, Q == &C } | |
152 | B = 4; Q = P; | |
153 | P = &B D = *Q; | |
154 | ||
155 | There is an obvious data dependency here, as the value loaded into D depends on | |
156 | the address retrieved from P by CPU 2. At the end of the sequence, any of the | |
157 | following results are possible: | |
158 | ||
159 | (Q == &A) and (D == 1) | |
160 | (Q == &B) and (D == 2) | |
161 | (Q == &B) and (D == 4) | |
162 | ||
163 | Note that CPU 2 will never try and load C into D because the CPU will load P | |
164 | into Q before issuing the load of *Q. | |
165 | ||
166 | ||
167 | DEVICE OPERATIONS | |
168 | ----------------- | |
169 | ||
170 | Some devices present their control interfaces as collections of memory | |
171 | locations, but the order in which the control registers are accessed is very | |
172 | important. For instance, imagine an ethernet card with a set of internal | |
173 | registers that are accessed through an address port register (A) and a data | |
174 | port register (D). To read internal register 5, the following code might then | |
175 | be used: | |
176 | ||
177 | *A = 5; | |
178 | x = *D; | |
179 | ||
180 | but this might show up as either of the following two sequences: | |
181 | ||
182 | STORE *A = 5, x = LOAD *D | |
183 | x = LOAD *D, STORE *A = 5 | |
184 | ||
185 | the second of which will almost certainly result in a malfunction, since it set | |
186 | the address _after_ attempting to read the register. | |
187 | ||
188 | ||
189 | GUARANTEES | |
190 | ---------- | |
191 | ||
192 | There are some minimal guarantees that may be expected of a CPU: | |
193 | ||
194 | (*) On any given CPU, dependent memory accesses will be issued in order, with | |
195 | respect to itself. This means that for: | |
196 | ||
2ecf8101 | 197 | ACCESS_ONCE(Q) = P; smp_read_barrier_depends(); D = ACCESS_ONCE(*Q); |
108b42b4 DH |
198 | |
199 | the CPU will issue the following memory operations: | |
200 | ||
201 | Q = LOAD P, D = LOAD *Q | |
202 | ||
2ecf8101 PM |
203 | and always in that order. On most systems, smp_read_barrier_depends() |
204 | does nothing, but it is required for DEC Alpha. The ACCESS_ONCE() | |
205 | is required to prevent compiler mischief. Please note that you | |
206 | should normally use something like rcu_dereference() instead of | |
207 | open-coding smp_read_barrier_depends(). | |
108b42b4 DH |
208 | |
209 | (*) Overlapping loads and stores within a particular CPU will appear to be | |
210 | ordered within that CPU. This means that for: | |
211 | ||
2ecf8101 | 212 | a = ACCESS_ONCE(*X); ACCESS_ONCE(*X) = b; |
108b42b4 DH |
213 | |
214 | the CPU will only issue the following sequence of memory operations: | |
215 | ||
216 | a = LOAD *X, STORE *X = b | |
217 | ||
218 | And for: | |
219 | ||
2ecf8101 | 220 | ACCESS_ONCE(*X) = c; d = ACCESS_ONCE(*X); |
108b42b4 DH |
221 | |
222 | the CPU will only issue: | |
223 | ||
224 | STORE *X = c, d = LOAD *X | |
225 | ||
fa00e7e1 | 226 | (Loads and stores overlap if they are targeted at overlapping pieces of |
108b42b4 DH |
227 | memory). |
228 | ||
229 | And there are a number of things that _must_ or _must_not_ be assumed: | |
230 | ||
2ecf8101 PM |
231 | (*) It _must_not_ be assumed that the compiler will do what you want with |
232 | memory references that are not protected by ACCESS_ONCE(). Without | |
233 | ACCESS_ONCE(), the compiler is within its rights to do all sorts | |
234 | of "creative" transformations: | |
235 | ||
236 | (-) Repeat the load, possibly getting a different value on the second | |
237 | and subsequent loads. This is especially prone to happen when | |
238 | register pressure is high. | |
239 | ||
240 | (-) Merge adjacent loads and stores to the same location. The most | |
241 | familiar example is the transformation from: | |
242 | ||
243 | while (a) | |
244 | do_something(); | |
245 | ||
246 | to something like: | |
247 | ||
248 | if (a) | |
249 | for (;;) | |
250 | do_something(); | |
251 | ||
252 | Using ACCESS_ONCE() as follows prevents this sort of optimization: | |
253 | ||
254 | while (ACCESS_ONCE(a)) | |
255 | do_something(); | |
256 | ||
257 | (-) "Store tearing", where a single store in the source code is split | |
258 | into smaller stores in the object code. Note that gcc really | |
259 | will do this on some architectures when storing certain constants. | |
260 | It can be cheaper to do a series of immediate stores than to | |
261 | form the constant in a register and then to store that register. | |
262 | ||
263 | (-) "Load tearing", which splits loads in a manner analogous to | |
264 | store tearing. | |
265 | ||
108b42b4 DH |
266 | (*) It _must_not_ be assumed that independent loads and stores will be issued |
267 | in the order given. This means that for: | |
268 | ||
269 | X = *A; Y = *B; *D = Z; | |
270 | ||
271 | we may get any of the following sequences: | |
272 | ||
273 | X = LOAD *A, Y = LOAD *B, STORE *D = Z | |
274 | X = LOAD *A, STORE *D = Z, Y = LOAD *B | |
275 | Y = LOAD *B, X = LOAD *A, STORE *D = Z | |
276 | Y = LOAD *B, STORE *D = Z, X = LOAD *A | |
277 | STORE *D = Z, X = LOAD *A, Y = LOAD *B | |
278 | STORE *D = Z, Y = LOAD *B, X = LOAD *A | |
279 | ||
280 | (*) It _must_ be assumed that overlapping memory accesses may be merged or | |
281 | discarded. This means that for: | |
282 | ||
283 | X = *A; Y = *(A + 4); | |
284 | ||
285 | we may get any one of the following sequences: | |
286 | ||
287 | X = LOAD *A; Y = LOAD *(A + 4); | |
288 | Y = LOAD *(A + 4); X = LOAD *A; | |
289 | {X, Y} = LOAD {*A, *(A + 4) }; | |
290 | ||
291 | And for: | |
292 | ||
f191eec5 | 293 | *A = X; *(A + 4) = Y; |
108b42b4 | 294 | |
f191eec5 | 295 | we may get any of: |
108b42b4 | 296 | |
f191eec5 PM |
297 | STORE *A = X; STORE *(A + 4) = Y; |
298 | STORE *(A + 4) = Y; STORE *A = X; | |
299 | STORE {*A, *(A + 4) } = {X, Y}; | |
108b42b4 DH |
300 | |
301 | ||
302 | ========================= | |
303 | WHAT ARE MEMORY BARRIERS? | |
304 | ========================= | |
305 | ||
306 | As can be seen above, independent memory operations are effectively performed | |
307 | in random order, but this can be a problem for CPU-CPU interaction and for I/O. | |
308 | What is required is some way of intervening to instruct the compiler and the | |
309 | CPU to restrict the order. | |
310 | ||
311 | Memory barriers are such interventions. They impose a perceived partial | |
2b94895b DH |
312 | ordering over the memory operations on either side of the barrier. |
313 | ||
314 | Such enforcement is important because the CPUs and other devices in a system | |
81fc6323 | 315 | can use a variety of tricks to improve performance, including reordering, |
2b94895b DH |
316 | deferral and combination of memory operations; speculative loads; speculative |
317 | branch prediction and various types of caching. Memory barriers are used to | |
318 | override or suppress these tricks, allowing the code to sanely control the | |
319 | interaction of multiple CPUs and/or devices. | |
108b42b4 DH |
320 | |
321 | ||
322 | VARIETIES OF MEMORY BARRIER | |
323 | --------------------------- | |
324 | ||
325 | Memory barriers come in four basic varieties: | |
326 | ||
327 | (1) Write (or store) memory barriers. | |
328 | ||
329 | A write memory barrier gives a guarantee that all the STORE operations | |
330 | specified before the barrier will appear to happen before all the STORE | |
331 | operations specified after the barrier with respect to the other | |
332 | components of the system. | |
333 | ||
334 | A write barrier is a partial ordering on stores only; it is not required | |
335 | to have any effect on loads. | |
336 | ||
6bc39274 | 337 | A CPU can be viewed as committing a sequence of store operations to the |
108b42b4 DH |
338 | memory system as time progresses. All stores before a write barrier will |
339 | occur in the sequence _before_ all the stores after the write barrier. | |
340 | ||
341 | [!] Note that write barriers should normally be paired with read or data | |
342 | dependency barriers; see the "SMP barrier pairing" subsection. | |
343 | ||
344 | ||
345 | (2) Data dependency barriers. | |
346 | ||
347 | A data dependency barrier is a weaker form of read barrier. In the case | |
348 | where two loads are performed such that the second depends on the result | |
349 | of the first (eg: the first load retrieves the address to which the second | |
350 | load will be directed), a data dependency barrier would be required to | |
351 | make sure that the target of the second load is updated before the address | |
352 | obtained by the first load is accessed. | |
353 | ||
354 | A data dependency barrier is a partial ordering on interdependent loads | |
355 | only; it is not required to have any effect on stores, independent loads | |
356 | or overlapping loads. | |
357 | ||
358 | As mentioned in (1), the other CPUs in the system can be viewed as | |
359 | committing sequences of stores to the memory system that the CPU being | |
360 | considered can then perceive. A data dependency barrier issued by the CPU | |
361 | under consideration guarantees that for any load preceding it, if that | |
362 | load touches one of a sequence of stores from another CPU, then by the | |
363 | time the barrier completes, the effects of all the stores prior to that | |
364 | touched by the load will be perceptible to any loads issued after the data | |
365 | dependency barrier. | |
366 | ||
367 | See the "Examples of memory barrier sequences" subsection for diagrams | |
368 | showing the ordering constraints. | |
369 | ||
370 | [!] Note that the first load really has to have a _data_ dependency and | |
371 | not a control dependency. If the address for the second load is dependent | |
372 | on the first load, but the dependency is through a conditional rather than | |
373 | actually loading the address itself, then it's a _control_ dependency and | |
374 | a full read barrier or better is required. See the "Control dependencies" | |
375 | subsection for more information. | |
376 | ||
377 | [!] Note that data dependency barriers should normally be paired with | |
378 | write barriers; see the "SMP barrier pairing" subsection. | |
379 | ||
380 | ||
381 | (3) Read (or load) memory barriers. | |
382 | ||
383 | A read barrier is a data dependency barrier plus a guarantee that all the | |
384 | LOAD operations specified before the barrier will appear to happen before | |
385 | all the LOAD operations specified after the barrier with respect to the | |
386 | other components of the system. | |
387 | ||
388 | A read barrier is a partial ordering on loads only; it is not required to | |
389 | have any effect on stores. | |
390 | ||
391 | Read memory barriers imply data dependency barriers, and so can substitute | |
392 | for them. | |
393 | ||
394 | [!] Note that read barriers should normally be paired with write barriers; | |
395 | see the "SMP barrier pairing" subsection. | |
396 | ||
397 | ||
398 | (4) General memory barriers. | |
399 | ||
670bd95e DH |
400 | A general memory barrier gives a guarantee that all the LOAD and STORE |
401 | operations specified before the barrier will appear to happen before all | |
402 | the LOAD and STORE operations specified after the barrier with respect to | |
403 | the other components of the system. | |
404 | ||
405 | A general memory barrier is a partial ordering over both loads and stores. | |
108b42b4 DH |
406 | |
407 | General memory barriers imply both read and write memory barriers, and so | |
408 | can substitute for either. | |
409 | ||
410 | ||
411 | And a couple of implicit varieties: | |
412 | ||
413 | (5) LOCK operations. | |
414 | ||
415 | This acts as a one-way permeable barrier. It guarantees that all memory | |
416 | operations after the LOCK operation will appear to happen after the LOCK | |
417 | operation with respect to the other components of the system. | |
418 | ||
419 | Memory operations that occur before a LOCK operation may appear to happen | |
420 | after it completes. | |
421 | ||
422 | A LOCK operation should almost always be paired with an UNLOCK operation. | |
423 | ||
424 | ||
425 | (6) UNLOCK operations. | |
426 | ||
427 | This also acts as a one-way permeable barrier. It guarantees that all | |
428 | memory operations before the UNLOCK operation will appear to happen before | |
429 | the UNLOCK operation with respect to the other components of the system. | |
430 | ||
431 | Memory operations that occur after an UNLOCK operation may appear to | |
432 | happen before it completes. | |
433 | ||
434 | LOCK and UNLOCK operations are guaranteed to appear with respect to each | |
435 | other strictly in the order specified. | |
436 | ||
437 | The use of LOCK and UNLOCK operations generally precludes the need for | |
438 | other sorts of memory barrier (but note the exceptions mentioned in the | |
439 | subsection "MMIO write barrier"). | |
440 | ||
441 | ||
442 | Memory barriers are only required where there's a possibility of interaction | |
443 | between two CPUs or between a CPU and a device. If it can be guaranteed that | |
444 | there won't be any such interaction in any particular piece of code, then | |
445 | memory barriers are unnecessary in that piece of code. | |
446 | ||
447 | ||
448 | Note that these are the _minimum_ guarantees. Different architectures may give | |
449 | more substantial guarantees, but they may _not_ be relied upon outside of arch | |
450 | specific code. | |
451 | ||
452 | ||
453 | WHAT MAY NOT BE ASSUMED ABOUT MEMORY BARRIERS? | |
454 | ---------------------------------------------- | |
455 | ||
456 | There are certain things that the Linux kernel memory barriers do not guarantee: | |
457 | ||
458 | (*) There is no guarantee that any of the memory accesses specified before a | |
459 | memory barrier will be _complete_ by the completion of a memory barrier | |
460 | instruction; the barrier can be considered to draw a line in that CPU's | |
461 | access queue that accesses of the appropriate type may not cross. | |
462 | ||
463 | (*) There is no guarantee that issuing a memory barrier on one CPU will have | |
464 | any direct effect on another CPU or any other hardware in the system. The | |
465 | indirect effect will be the order in which the second CPU sees the effects | |
466 | of the first CPU's accesses occur, but see the next point: | |
467 | ||
6bc39274 | 468 | (*) There is no guarantee that a CPU will see the correct order of effects |
108b42b4 DH |
469 | from a second CPU's accesses, even _if_ the second CPU uses a memory |
470 | barrier, unless the first CPU _also_ uses a matching memory barrier (see | |
471 | the subsection on "SMP Barrier Pairing"). | |
472 | ||
473 | (*) There is no guarantee that some intervening piece of off-the-CPU | |
474 | hardware[*] will not reorder the memory accesses. CPU cache coherency | |
475 | mechanisms should propagate the indirect effects of a memory barrier | |
476 | between CPUs, but might not do so in order. | |
477 | ||
478 | [*] For information on bus mastering DMA and coherency please read: | |
479 | ||
4b5ff469 | 480 | Documentation/PCI/pci.txt |
395cf969 | 481 | Documentation/DMA-API-HOWTO.txt |
108b42b4 DH |
482 | Documentation/DMA-API.txt |
483 | ||
484 | ||
485 | DATA DEPENDENCY BARRIERS | |
486 | ------------------------ | |
487 | ||
488 | The usage requirements of data dependency barriers are a little subtle, and | |
489 | it's not always obvious that they're needed. To illustrate, consider the | |
490 | following sequence of events: | |
491 | ||
2ecf8101 PM |
492 | CPU 1 CPU 2 |
493 | =============== =============== | |
108b42b4 DH |
494 | { A == 1, B == 2, C = 3, P == &A, Q == &C } |
495 | B = 4; | |
496 | <write barrier> | |
2ecf8101 PM |
497 | ACCESS_ONCE(P) = &B |
498 | Q = ACCESS_ONCE(P); | |
499 | D = *Q; | |
108b42b4 DH |
500 | |
501 | There's a clear data dependency here, and it would seem that by the end of the | |
502 | sequence, Q must be either &A or &B, and that: | |
503 | ||
504 | (Q == &A) implies (D == 1) | |
505 | (Q == &B) implies (D == 4) | |
506 | ||
81fc6323 | 507 | But! CPU 2's perception of P may be updated _before_ its perception of B, thus |
108b42b4 DH |
508 | leading to the following situation: |
509 | ||
510 | (Q == &B) and (D == 2) ???? | |
511 | ||
512 | Whilst this may seem like a failure of coherency or causality maintenance, it | |
513 | isn't, and this behaviour can be observed on certain real CPUs (such as the DEC | |
514 | Alpha). | |
515 | ||
2b94895b DH |
516 | To deal with this, a data dependency barrier or better must be inserted |
517 | between the address load and the data load: | |
108b42b4 | 518 | |
2ecf8101 PM |
519 | CPU 1 CPU 2 |
520 | =============== =============== | |
108b42b4 DH |
521 | { A == 1, B == 2, C = 3, P == &A, Q == &C } |
522 | B = 4; | |
523 | <write barrier> | |
2ecf8101 PM |
524 | ACCESS_ONCE(P) = &B |
525 | Q = ACCESS_ONCE(P); | |
526 | <data dependency barrier> | |
527 | D = *Q; | |
108b42b4 DH |
528 | |
529 | This enforces the occurrence of one of the two implications, and prevents the | |
530 | third possibility from arising. | |
531 | ||
532 | [!] Note that this extremely counterintuitive situation arises most easily on | |
533 | machines with split caches, so that, for example, one cache bank processes | |
534 | even-numbered cache lines and the other bank processes odd-numbered cache | |
535 | lines. The pointer P might be stored in an odd-numbered cache line, and the | |
536 | variable B might be stored in an even-numbered cache line. Then, if the | |
537 | even-numbered bank of the reading CPU's cache is extremely busy while the | |
538 | odd-numbered bank is idle, one can see the new value of the pointer P (&B), | |
6bc39274 | 539 | but the old value of the variable B (2). |
108b42b4 DH |
540 | |
541 | ||
e0edc78f | 542 | Another example of where data dependency barriers might be required is where a |
108b42b4 DH |
543 | number is read from memory and then used to calculate the index for an array |
544 | access: | |
545 | ||
2ecf8101 PM |
546 | CPU 1 CPU 2 |
547 | =============== =============== | |
108b42b4 DH |
548 | { M[0] == 1, M[1] == 2, M[3] = 3, P == 0, Q == 3 } |
549 | M[1] = 4; | |
550 | <write barrier> | |
2ecf8101 PM |
551 | ACCESS_ONCE(P) = 1 |
552 | Q = ACCESS_ONCE(P); | |
553 | <data dependency barrier> | |
554 | D = M[Q]; | |
108b42b4 DH |
555 | |
556 | ||
2ecf8101 PM |
557 | The data dependency barrier is very important to the RCU system, |
558 | for example. See rcu_assign_pointer() and rcu_dereference() in | |
559 | include/linux/rcupdate.h. This permits the current target of an RCU'd | |
560 | pointer to be replaced with a new modified target, without the replacement | |
561 | target appearing to be incompletely initialised. | |
108b42b4 DH |
562 | |
563 | See also the subsection on "Cache Coherency" for a more thorough example. | |
564 | ||
565 | ||
566 | CONTROL DEPENDENCIES | |
567 | -------------------- | |
568 | ||
569 | A control dependency requires a full read memory barrier, not simply a data | |
570 | dependency barrier to make it work correctly. Consider the following bit of | |
571 | code: | |
572 | ||
2ecf8101 | 573 | q = ACCESS_ONCE(a); |
45c8a36a PM |
574 | if (p) { |
575 | <data dependency barrier> | |
2ecf8101 | 576 | q = ACCESS_ONCE(b); |
45c8a36a | 577 | } |
108b42b4 DH |
578 | x = *q; |
579 | ||
580 | This will not have the desired effect because there is no actual data | |
2ecf8101 PM |
581 | dependency, but rather a control dependency that the CPU may short-circuit |
582 | by attempting to predict the outcome in advance, so that other CPUs see | |
583 | the load from b as having happened before the load from a. In such a | |
584 | case what's actually required is: | |
108b42b4 | 585 | |
2ecf8101 | 586 | q = ACCESS_ONCE(a); |
45c8a36a PM |
587 | if (p) { |
588 | <read barrier> | |
2ecf8101 | 589 | q = ACCESS_ONCE(b); |
45c8a36a | 590 | } |
108b42b4 DH |
591 | x = *q; |
592 | ||
593 | ||
594 | SMP BARRIER PAIRING | |
595 | ------------------- | |
596 | ||
597 | When dealing with CPU-CPU interactions, certain types of memory barrier should | |
598 | always be paired. A lack of appropriate pairing is almost certainly an error. | |
599 | ||
600 | A write barrier should always be paired with a data dependency barrier or read | |
601 | barrier, though a general barrier would also be viable. Similarly a read | |
602 | barrier or a data dependency barrier should always be paired with at least an | |
603 | write barrier, though, again, a general barrier is viable: | |
604 | ||
2ecf8101 PM |
605 | CPU 1 CPU 2 |
606 | =============== =============== | |
607 | ACCESS_ONCE(a) = 1; | |
108b42b4 | 608 | <write barrier> |
2ecf8101 PM |
609 | ACCESS_ONCE(b) = 2; x = ACCESS_ONCE(b); |
610 | <read barrier> | |
611 | y = ACCESS_ONCE(a); | |
108b42b4 DH |
612 | |
613 | Or: | |
614 | ||
2ecf8101 PM |
615 | CPU 1 CPU 2 |
616 | =============== =============================== | |
108b42b4 DH |
617 | a = 1; |
618 | <write barrier> | |
2ecf8101 PM |
619 | ACCESS_ONCE(b) = &a; x = ACCESS_ONCE(b); |
620 | <data dependency barrier> | |
621 | y = *x; | |
108b42b4 DH |
622 | |
623 | Basically, the read barrier always has to be there, even though it can be of | |
624 | the "weaker" type. | |
625 | ||
670bd95e | 626 | [!] Note that the stores before the write barrier would normally be expected to |
81fc6323 | 627 | match the loads after the read barrier or the data dependency barrier, and vice |
670bd95e DH |
628 | versa: |
629 | ||
2ecf8101 PM |
630 | CPU 1 CPU 2 |
631 | =================== =================== | |
632 | ACCESS_ONCE(a) = 1; }---- --->{ v = ACCESS_ONCE(c); | |
633 | ACCESS_ONCE(b) = 2; } \ / { w = ACCESS_ONCE(d); | |
634 | <write barrier> \ <read barrier> | |
635 | ACCESS_ONCE(c) = 3; } / \ { x = ACCESS_ONCE(a); | |
636 | ACCESS_ONCE(d) = 4; }---- --->{ y = ACCESS_ONCE(b); | |
670bd95e | 637 | |
108b42b4 DH |
638 | |
639 | EXAMPLES OF MEMORY BARRIER SEQUENCES | |
640 | ------------------------------------ | |
641 | ||
81fc6323 | 642 | Firstly, write barriers act as partial orderings on store operations. |
108b42b4 DH |
643 | Consider the following sequence of events: |
644 | ||
645 | CPU 1 | |
646 | ======================= | |
647 | STORE A = 1 | |
648 | STORE B = 2 | |
649 | STORE C = 3 | |
650 | <write barrier> | |
651 | STORE D = 4 | |
652 | STORE E = 5 | |
653 | ||
654 | This sequence of events is committed to the memory coherence system in an order | |
655 | that the rest of the system might perceive as the unordered set of { STORE A, | |
80f7228b | 656 | STORE B, STORE C } all occurring before the unordered set of { STORE D, STORE E |
108b42b4 DH |
657 | }: |
658 | ||
659 | +-------+ : : | |
660 | | | +------+ | |
661 | | |------>| C=3 | } /\ | |
81fc6323 JP |
662 | | | : +------+ }----- \ -----> Events perceptible to |
663 | | | : | A=1 | } \/ the rest of the system | |
108b42b4 DH |
664 | | | : +------+ } |
665 | | CPU 1 | : | B=2 | } | |
666 | | | +------+ } | |
667 | | | wwwwwwwwwwwwwwww } <--- At this point the write barrier | |
668 | | | +------+ } requires all stores prior to the | |
669 | | | : | E=5 | } barrier to be committed before | |
81fc6323 | 670 | | | : +------+ } further stores may take place |
108b42b4 DH |
671 | | |------>| D=4 | } |
672 | | | +------+ | |
673 | +-------+ : : | |
674 | | | |
670bd95e DH |
675 | | Sequence in which stores are committed to the |
676 | | memory system by CPU 1 | |
108b42b4 DH |
677 | V |
678 | ||
679 | ||
81fc6323 | 680 | Secondly, data dependency barriers act as partial orderings on data-dependent |
108b42b4 DH |
681 | loads. Consider the following sequence of events: |
682 | ||
683 | CPU 1 CPU 2 | |
684 | ======================= ======================= | |
c14038c3 | 685 | { B = 7; X = 9; Y = 8; C = &Y } |
108b42b4 DH |
686 | STORE A = 1 |
687 | STORE B = 2 | |
688 | <write barrier> | |
689 | STORE C = &B LOAD X | |
690 | STORE D = 4 LOAD C (gets &B) | |
691 | LOAD *C (reads B) | |
692 | ||
693 | Without intervention, CPU 2 may perceive the events on CPU 1 in some | |
694 | effectively random order, despite the write barrier issued by CPU 1: | |
695 | ||
696 | +-------+ : : : : | |
697 | | | +------+ +-------+ | Sequence of update | |
698 | | |------>| B=2 |----- --->| Y->8 | | of perception on | |
699 | | | : +------+ \ +-------+ | CPU 2 | |
700 | | CPU 1 | : | A=1 | \ --->| C->&Y | V | |
701 | | | +------+ | +-------+ | |
702 | | | wwwwwwwwwwwwwwww | : : | |
703 | | | +------+ | : : | |
704 | | | : | C=&B |--- | : : +-------+ | |
705 | | | : +------+ \ | +-------+ | | | |
706 | | |------>| D=4 | ----------->| C->&B |------>| | | |
707 | | | +------+ | +-------+ | | | |
708 | +-------+ : : | : : | | | |
709 | | : : | | | |
710 | | : : | CPU 2 | | |
711 | | +-------+ | | | |
712 | Apparently incorrect ---> | | B->7 |------>| | | |
713 | perception of B (!) | +-------+ | | | |
714 | | : : | | | |
715 | | +-------+ | | | |
716 | The load of X holds ---> \ | X->9 |------>| | | |
717 | up the maintenance \ +-------+ | | | |
718 | of coherence of B ----->| B->2 | +-------+ | |
719 | +-------+ | |
720 | : : | |
721 | ||
722 | ||
723 | In the above example, CPU 2 perceives that B is 7, despite the load of *C | |
670e9f34 | 724 | (which would be B) coming after the LOAD of C. |
108b42b4 DH |
725 | |
726 | If, however, a data dependency barrier were to be placed between the load of C | |
c14038c3 DH |
727 | and the load of *C (ie: B) on CPU 2: |
728 | ||
729 | CPU 1 CPU 2 | |
730 | ======================= ======================= | |
731 | { B = 7; X = 9; Y = 8; C = &Y } | |
732 | STORE A = 1 | |
733 | STORE B = 2 | |
734 | <write barrier> | |
735 | STORE C = &B LOAD X | |
736 | STORE D = 4 LOAD C (gets &B) | |
737 | <data dependency barrier> | |
738 | LOAD *C (reads B) | |
739 | ||
740 | then the following will occur: | |
108b42b4 DH |
741 | |
742 | +-------+ : : : : | |
743 | | | +------+ +-------+ | |
744 | | |------>| B=2 |----- --->| Y->8 | | |
745 | | | : +------+ \ +-------+ | |
746 | | CPU 1 | : | A=1 | \ --->| C->&Y | | |
747 | | | +------+ | +-------+ | |
748 | | | wwwwwwwwwwwwwwww | : : | |
749 | | | +------+ | : : | |
750 | | | : | C=&B |--- | : : +-------+ | |
751 | | | : +------+ \ | +-------+ | | | |
752 | | |------>| D=4 | ----------->| C->&B |------>| | | |
753 | | | +------+ | +-------+ | | | |
754 | +-------+ : : | : : | | | |
755 | | : : | | | |
756 | | : : | CPU 2 | | |
757 | | +-------+ | | | |
670bd95e DH |
758 | | | X->9 |------>| | |
759 | | +-------+ | | | |
760 | Makes sure all effects ---> \ ddddddddddddddddd | | | |
761 | prior to the store of C \ +-------+ | | | |
762 | are perceptible to ----->| B->2 |------>| | | |
763 | subsequent loads +-------+ | | | |
108b42b4 DH |
764 | : : +-------+ |
765 | ||
766 | ||
767 | And thirdly, a read barrier acts as a partial order on loads. Consider the | |
768 | following sequence of events: | |
769 | ||
770 | CPU 1 CPU 2 | |
771 | ======================= ======================= | |
670bd95e | 772 | { A = 0, B = 9 } |
108b42b4 | 773 | STORE A=1 |
108b42b4 | 774 | <write barrier> |
670bd95e | 775 | STORE B=2 |
108b42b4 | 776 | LOAD B |
670bd95e | 777 | LOAD A |
108b42b4 DH |
778 | |
779 | Without intervention, CPU 2 may then choose to perceive the events on CPU 1 in | |
780 | some effectively random order, despite the write barrier issued by CPU 1: | |
781 | ||
670bd95e DH |
782 | +-------+ : : : : |
783 | | | +------+ +-------+ | |
784 | | |------>| A=1 |------ --->| A->0 | | |
785 | | | +------+ \ +-------+ | |
786 | | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 | | |
787 | | | +------+ | +-------+ | |
788 | | |------>| B=2 |--- | : : | |
789 | | | +------+ \ | : : +-------+ | |
790 | +-------+ : : \ | +-------+ | | | |
791 | ---------->| B->2 |------>| | | |
792 | | +-------+ | CPU 2 | | |
793 | | | A->0 |------>| | | |
794 | | +-------+ | | | |
795 | | : : +-------+ | |
796 | \ : : | |
797 | \ +-------+ | |
798 | ---->| A->1 | | |
799 | +-------+ | |
800 | : : | |
108b42b4 | 801 | |
670bd95e | 802 | |
6bc39274 | 803 | If, however, a read barrier were to be placed between the load of B and the |
670bd95e DH |
804 | load of A on CPU 2: |
805 | ||
806 | CPU 1 CPU 2 | |
807 | ======================= ======================= | |
808 | { A = 0, B = 9 } | |
809 | STORE A=1 | |
810 | <write barrier> | |
811 | STORE B=2 | |
812 | LOAD B | |
813 | <read barrier> | |
814 | LOAD A | |
815 | ||
816 | then the partial ordering imposed by CPU 1 will be perceived correctly by CPU | |
817 | 2: | |
818 | ||
819 | +-------+ : : : : | |
820 | | | +------+ +-------+ | |
821 | | |------>| A=1 |------ --->| A->0 | | |
822 | | | +------+ \ +-------+ | |
823 | | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 | | |
824 | | | +------+ | +-------+ | |
825 | | |------>| B=2 |--- | : : | |
826 | | | +------+ \ | : : +-------+ | |
827 | +-------+ : : \ | +-------+ | | | |
828 | ---------->| B->2 |------>| | | |
829 | | +-------+ | CPU 2 | | |
830 | | : : | | | |
831 | | : : | | | |
832 | At this point the read ----> \ rrrrrrrrrrrrrrrrr | | | |
833 | barrier causes all effects \ +-------+ | | | |
834 | prior to the storage of B ---->| A->1 |------>| | | |
835 | to be perceptible to CPU 2 +-------+ | | | |
836 | : : +-------+ | |
837 | ||
838 | ||
839 | To illustrate this more completely, consider what could happen if the code | |
840 | contained a load of A either side of the read barrier: | |
841 | ||
842 | CPU 1 CPU 2 | |
843 | ======================= ======================= | |
844 | { A = 0, B = 9 } | |
845 | STORE A=1 | |
846 | <write barrier> | |
847 | STORE B=2 | |
848 | LOAD B | |
849 | LOAD A [first load of A] | |
850 | <read barrier> | |
851 | LOAD A [second load of A] | |
852 | ||
853 | Even though the two loads of A both occur after the load of B, they may both | |
854 | come up with different values: | |
855 | ||
856 | +-------+ : : : : | |
857 | | | +------+ +-------+ | |
858 | | |------>| A=1 |------ --->| A->0 | | |
859 | | | +------+ \ +-------+ | |
860 | | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 | | |
861 | | | +------+ | +-------+ | |
862 | | |------>| B=2 |--- | : : | |
863 | | | +------+ \ | : : +-------+ | |
864 | +-------+ : : \ | +-------+ | | | |
865 | ---------->| B->2 |------>| | | |
866 | | +-------+ | CPU 2 | | |
867 | | : : | | | |
868 | | : : | | | |
869 | | +-------+ | | | |
870 | | | A->0 |------>| 1st | | |
871 | | +-------+ | | | |
872 | At this point the read ----> \ rrrrrrrrrrrrrrrrr | | | |
873 | barrier causes all effects \ +-------+ | | | |
874 | prior to the storage of B ---->| A->1 |------>| 2nd | | |
875 | to be perceptible to CPU 2 +-------+ | | | |
876 | : : +-------+ | |
877 | ||
878 | ||
879 | But it may be that the update to A from CPU 1 becomes perceptible to CPU 2 | |
880 | before the read barrier completes anyway: | |
881 | ||
882 | +-------+ : : : : | |
883 | | | +------+ +-------+ | |
884 | | |------>| A=1 |------ --->| A->0 | | |
885 | | | +------+ \ +-------+ | |
886 | | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 | | |
887 | | | +------+ | +-------+ | |
888 | | |------>| B=2 |--- | : : | |
889 | | | +------+ \ | : : +-------+ | |
890 | +-------+ : : \ | +-------+ | | | |
891 | ---------->| B->2 |------>| | | |
892 | | +-------+ | CPU 2 | | |
893 | | : : | | | |
894 | \ : : | | | |
895 | \ +-------+ | | | |
896 | ---->| A->1 |------>| 1st | | |
897 | +-------+ | | | |
898 | rrrrrrrrrrrrrrrrr | | | |
899 | +-------+ | | | |
900 | | A->1 |------>| 2nd | | |
901 | +-------+ | | | |
902 | : : +-------+ | |
903 | ||
904 | ||
905 | The guarantee is that the second load will always come up with A == 1 if the | |
906 | load of B came up with B == 2. No such guarantee exists for the first load of | |
907 | A; that may come up with either A == 0 or A == 1. | |
908 | ||
909 | ||
910 | READ MEMORY BARRIERS VS LOAD SPECULATION | |
911 | ---------------------------------------- | |
912 | ||
913 | Many CPUs speculate with loads: that is they see that they will need to load an | |
914 | item from memory, and they find a time where they're not using the bus for any | |
915 | other loads, and so do the load in advance - even though they haven't actually | |
916 | got to that point in the instruction execution flow yet. This permits the | |
917 | actual load instruction to potentially complete immediately because the CPU | |
918 | already has the value to hand. | |
919 | ||
920 | It may turn out that the CPU didn't actually need the value - perhaps because a | |
921 | branch circumvented the load - in which case it can discard the value or just | |
922 | cache it for later use. | |
923 | ||
924 | Consider: | |
925 | ||
e0edc78f | 926 | CPU 1 CPU 2 |
670bd95e | 927 | ======================= ======================= |
e0edc78f IM |
928 | LOAD B |
929 | DIVIDE } Divide instructions generally | |
930 | DIVIDE } take a long time to perform | |
931 | LOAD A | |
670bd95e DH |
932 | |
933 | Which might appear as this: | |
934 | ||
935 | : : +-------+ | |
936 | +-------+ | | | |
937 | --->| B->2 |------>| | | |
938 | +-------+ | CPU 2 | | |
939 | : :DIVIDE | | | |
940 | +-------+ | | | |
941 | The CPU being busy doing a ---> --->| A->0 |~~~~ | | | |
942 | division speculates on the +-------+ ~ | | | |
943 | LOAD of A : : ~ | | | |
944 | : :DIVIDE | | | |
945 | : : ~ | | | |
946 | Once the divisions are complete --> : : ~-->| | | |
947 | the CPU can then perform the : : | | | |
948 | LOAD with immediate effect : : +-------+ | |
949 | ||
950 | ||
951 | Placing a read barrier or a data dependency barrier just before the second | |
952 | load: | |
953 | ||
e0edc78f | 954 | CPU 1 CPU 2 |
670bd95e | 955 | ======================= ======================= |
e0edc78f IM |
956 | LOAD B |
957 | DIVIDE | |
958 | DIVIDE | |
670bd95e | 959 | <read barrier> |
e0edc78f | 960 | LOAD A |
670bd95e DH |
961 | |
962 | will force any value speculatively obtained to be reconsidered to an extent | |
963 | dependent on the type of barrier used. If there was no change made to the | |
964 | speculated memory location, then the speculated value will just be used: | |
965 | ||
966 | : : +-------+ | |
967 | +-------+ | | | |
968 | --->| B->2 |------>| | | |
969 | +-------+ | CPU 2 | | |
970 | : :DIVIDE | | | |
971 | +-------+ | | | |
972 | The CPU being busy doing a ---> --->| A->0 |~~~~ | | | |
973 | division speculates on the +-------+ ~ | | | |
974 | LOAD of A : : ~ | | | |
975 | : :DIVIDE | | | |
976 | : : ~ | | | |
977 | : : ~ | | | |
978 | rrrrrrrrrrrrrrrr~ | | | |
979 | : : ~ | | | |
980 | : : ~-->| | | |
981 | : : | | | |
982 | : : +-------+ | |
983 | ||
984 | ||
985 | but if there was an update or an invalidation from another CPU pending, then | |
986 | the speculation will be cancelled and the value reloaded: | |
987 | ||
988 | : : +-------+ | |
989 | +-------+ | | | |
990 | --->| B->2 |------>| | | |
991 | +-------+ | CPU 2 | | |
992 | : :DIVIDE | | | |
993 | +-------+ | | | |
994 | The CPU being busy doing a ---> --->| A->0 |~~~~ | | | |
995 | division speculates on the +-------+ ~ | | | |
996 | LOAD of A : : ~ | | | |
997 | : :DIVIDE | | | |
998 | : : ~ | | | |
999 | : : ~ | | | |
1000 | rrrrrrrrrrrrrrrrr | | | |
1001 | +-------+ | | | |
1002 | The speculation is discarded ---> --->| A->1 |------>| | | |
1003 | and an updated value is +-------+ | | | |
1004 | retrieved : : +-------+ | |
108b42b4 DH |
1005 | |
1006 | ||
241e6663 PM |
1007 | TRANSITIVITY |
1008 | ------------ | |
1009 | ||
1010 | Transitivity is a deeply intuitive notion about ordering that is not | |
1011 | always provided by real computer systems. The following example | |
1012 | demonstrates transitivity (also called "cumulativity"): | |
1013 | ||
1014 | CPU 1 CPU 2 CPU 3 | |
1015 | ======================= ======================= ======================= | |
1016 | { X = 0, Y = 0 } | |
1017 | STORE X=1 LOAD X STORE Y=1 | |
1018 | <general barrier> <general barrier> | |
1019 | LOAD Y LOAD X | |
1020 | ||
1021 | Suppose that CPU 2's load from X returns 1 and its load from Y returns 0. | |
1022 | This indicates that CPU 2's load from X in some sense follows CPU 1's | |
1023 | store to X and that CPU 2's load from Y in some sense preceded CPU 3's | |
1024 | store to Y. The question is then "Can CPU 3's load from X return 0?" | |
1025 | ||
1026 | Because CPU 2's load from X in some sense came after CPU 1's store, it | |
1027 | is natural to expect that CPU 3's load from X must therefore return 1. | |
1028 | This expectation is an example of transitivity: if a load executing on | |
1029 | CPU A follows a load from the same variable executing on CPU B, then | |
1030 | CPU A's load must either return the same value that CPU B's load did, | |
1031 | or must return some later value. | |
1032 | ||
1033 | In the Linux kernel, use of general memory barriers guarantees | |
1034 | transitivity. Therefore, in the above example, if CPU 2's load from X | |
1035 | returns 1 and its load from Y returns 0, then CPU 3's load from X must | |
1036 | also return 1. | |
1037 | ||
1038 | However, transitivity is -not- guaranteed for read or write barriers. | |
1039 | For example, suppose that CPU 2's general barrier in the above example | |
1040 | is changed to a read barrier as shown below: | |
1041 | ||
1042 | CPU 1 CPU 2 CPU 3 | |
1043 | ======================= ======================= ======================= | |
1044 | { X = 0, Y = 0 } | |
1045 | STORE X=1 LOAD X STORE Y=1 | |
1046 | <read barrier> <general barrier> | |
1047 | LOAD Y LOAD X | |
1048 | ||
1049 | This substitution destroys transitivity: in this example, it is perfectly | |
1050 | legal for CPU 2's load from X to return 1, its load from Y to return 0, | |
1051 | and CPU 3's load from X to return 0. | |
1052 | ||
1053 | The key point is that although CPU 2's read barrier orders its pair | |
1054 | of loads, it does not guarantee to order CPU 1's store. Therefore, if | |
1055 | this example runs on a system where CPUs 1 and 2 share a store buffer | |
1056 | or a level of cache, CPU 2 might have early access to CPU 1's writes. | |
1057 | General barriers are therefore required to ensure that all CPUs agree | |
1058 | on the combined order of CPU 1's and CPU 2's accesses. | |
1059 | ||
1060 | To reiterate, if your code requires transitivity, use general barriers | |
1061 | throughout. | |
1062 | ||
1063 | ||
108b42b4 DH |
1064 | ======================== |
1065 | EXPLICIT KERNEL BARRIERS | |
1066 | ======================== | |
1067 | ||
1068 | The Linux kernel has a variety of different barriers that act at different | |
1069 | levels: | |
1070 | ||
1071 | (*) Compiler barrier. | |
1072 | ||
1073 | (*) CPU memory barriers. | |
1074 | ||
1075 | (*) MMIO write barrier. | |
1076 | ||
1077 | ||
1078 | COMPILER BARRIER | |
1079 | ---------------- | |
1080 | ||
1081 | The Linux kernel has an explicit compiler barrier function that prevents the | |
1082 | compiler from moving the memory accesses either side of it to the other side: | |
1083 | ||
1084 | barrier(); | |
1085 | ||
81fc6323 | 1086 | This is a general barrier - lesser varieties of compiler barrier do not exist. |
108b42b4 DH |
1087 | |
1088 | The compiler barrier has no direct effect on the CPU, which may then reorder | |
1089 | things however it wishes. | |
1090 | ||
1091 | ||
1092 | CPU MEMORY BARRIERS | |
1093 | ------------------- | |
1094 | ||
1095 | The Linux kernel has eight basic CPU memory barriers: | |
1096 | ||
1097 | TYPE MANDATORY SMP CONDITIONAL | |
1098 | =============== ======================= =========================== | |
1099 | GENERAL mb() smp_mb() | |
1100 | WRITE wmb() smp_wmb() | |
1101 | READ rmb() smp_rmb() | |
1102 | DATA DEPENDENCY read_barrier_depends() smp_read_barrier_depends() | |
1103 | ||
1104 | ||
73f10281 NP |
1105 | All memory barriers except the data dependency barriers imply a compiler |
1106 | barrier. Data dependencies do not impose any additional compiler ordering. | |
1107 | ||
1108 | Aside: In the case of data dependencies, the compiler would be expected to | |
1109 | issue the loads in the correct order (eg. `a[b]` would have to load the value | |
1110 | of b before loading a[b]), however there is no guarantee in the C specification | |
1111 | that the compiler may not speculate the value of b (eg. is equal to 1) and load | |
1112 | a before b (eg. tmp = a[1]; if (b != 1) tmp = a[b]; ). There is also the | |
1113 | problem of a compiler reloading b after having loaded a[b], thus having a newer | |
1114 | copy of b than a[b]. A consensus has not yet been reached about these problems, | |
1115 | however the ACCESS_ONCE macro is a good place to start looking. | |
108b42b4 DH |
1116 | |
1117 | SMP memory barriers are reduced to compiler barriers on uniprocessor compiled | |
81fc6323 | 1118 | systems because it is assumed that a CPU will appear to be self-consistent, |
108b42b4 DH |
1119 | and will order overlapping accesses correctly with respect to itself. |
1120 | ||
1121 | [!] Note that SMP memory barriers _must_ be used to control the ordering of | |
1122 | references to shared memory on SMP systems, though the use of locking instead | |
1123 | is sufficient. | |
1124 | ||
1125 | Mandatory barriers should not be used to control SMP effects, since mandatory | |
1126 | barriers unnecessarily impose overhead on UP systems. They may, however, be | |
1127 | used to control MMIO effects on accesses through relaxed memory I/O windows. | |
1128 | These are required even on non-SMP systems as they affect the order in which | |
1129 | memory operations appear to a device by prohibiting both the compiler and the | |
1130 | CPU from reordering them. | |
1131 | ||
1132 | ||
1133 | There are some more advanced barrier functions: | |
1134 | ||
1135 | (*) set_mb(var, value) | |
108b42b4 | 1136 | |
75b2bd55 | 1137 | This assigns the value to the variable and then inserts a full memory |
f92213ba | 1138 | barrier after it, depending on the function. It isn't guaranteed to |
108b42b4 DH |
1139 | insert anything more than a compiler barrier in a UP compilation. |
1140 | ||
1141 | ||
1142 | (*) smp_mb__before_atomic_dec(); | |
1143 | (*) smp_mb__after_atomic_dec(); | |
1144 | (*) smp_mb__before_atomic_inc(); | |
1145 | (*) smp_mb__after_atomic_inc(); | |
1146 | ||
1147 | These are for use with atomic add, subtract, increment and decrement | |
dbc8700e DH |
1148 | functions that don't return a value, especially when used for reference |
1149 | counting. These functions do not imply memory barriers. | |
108b42b4 DH |
1150 | |
1151 | As an example, consider a piece of code that marks an object as being dead | |
1152 | and then decrements the object's reference count: | |
1153 | ||
1154 | obj->dead = 1; | |
1155 | smp_mb__before_atomic_dec(); | |
1156 | atomic_dec(&obj->ref_count); | |
1157 | ||
1158 | This makes sure that the death mark on the object is perceived to be set | |
1159 | *before* the reference counter is decremented. | |
1160 | ||
1161 | See Documentation/atomic_ops.txt for more information. See the "Atomic | |
1162 | operations" subsection for information on where to use these. | |
1163 | ||
1164 | ||
1165 | (*) smp_mb__before_clear_bit(void); | |
1166 | (*) smp_mb__after_clear_bit(void); | |
1167 | ||
1168 | These are for use similar to the atomic inc/dec barriers. These are | |
1169 | typically used for bitwise unlocking operations, so care must be taken as | |
1170 | there are no implicit memory barriers here either. | |
1171 | ||
1172 | Consider implementing an unlock operation of some nature by clearing a | |
1173 | locking bit. The clear_bit() would then need to be barriered like this: | |
1174 | ||
1175 | smp_mb__before_clear_bit(); | |
1176 | clear_bit( ... ); | |
1177 | ||
1178 | This prevents memory operations before the clear leaking to after it. See | |
1179 | the subsection on "Locking Functions" with reference to UNLOCK operation | |
1180 | implications. | |
1181 | ||
1182 | See Documentation/atomic_ops.txt for more information. See the "Atomic | |
1183 | operations" subsection for information on where to use these. | |
1184 | ||
1185 | ||
1186 | MMIO WRITE BARRIER | |
1187 | ------------------ | |
1188 | ||
1189 | The Linux kernel also has a special barrier for use with memory-mapped I/O | |
1190 | writes: | |
1191 | ||
1192 | mmiowb(); | |
1193 | ||
1194 | This is a variation on the mandatory write barrier that causes writes to weakly | |
1195 | ordered I/O regions to be partially ordered. Its effects may go beyond the | |
1196 | CPU->Hardware interface and actually affect the hardware at some level. | |
1197 | ||
1198 | See the subsection "Locks vs I/O accesses" for more information. | |
1199 | ||
1200 | ||
1201 | =============================== | |
1202 | IMPLICIT KERNEL MEMORY BARRIERS | |
1203 | =============================== | |
1204 | ||
1205 | Some of the other functions in the linux kernel imply memory barriers, amongst | |
670bd95e | 1206 | which are locking and scheduling functions. |
108b42b4 DH |
1207 | |
1208 | This specification is a _minimum_ guarantee; any particular architecture may | |
1209 | provide more substantial guarantees, but these may not be relied upon outside | |
1210 | of arch specific code. | |
1211 | ||
1212 | ||
1213 | LOCKING FUNCTIONS | |
1214 | ----------------- | |
1215 | ||
1216 | The Linux kernel has a number of locking constructs: | |
1217 | ||
1218 | (*) spin locks | |
1219 | (*) R/W spin locks | |
1220 | (*) mutexes | |
1221 | (*) semaphores | |
1222 | (*) R/W semaphores | |
1223 | (*) RCU | |
1224 | ||
1225 | In all cases there are variants on "LOCK" operations and "UNLOCK" operations | |
1226 | for each construct. These operations all imply certain barriers: | |
1227 | ||
1228 | (1) LOCK operation implication: | |
1229 | ||
1230 | Memory operations issued after the LOCK will be completed after the LOCK | |
1231 | operation has completed. | |
1232 | ||
1233 | Memory operations issued before the LOCK may be completed after the LOCK | |
1234 | operation has completed. | |
1235 | ||
1236 | (2) UNLOCK operation implication: | |
1237 | ||
1238 | Memory operations issued before the UNLOCK will be completed before the | |
1239 | UNLOCK operation has completed. | |
1240 | ||
1241 | Memory operations issued after the UNLOCK may be completed before the | |
1242 | UNLOCK operation has completed. | |
1243 | ||
1244 | (3) LOCK vs LOCK implication: | |
1245 | ||
1246 | All LOCK operations issued before another LOCK operation will be completed | |
1247 | before that LOCK operation. | |
1248 | ||
1249 | (4) LOCK vs UNLOCK implication: | |
1250 | ||
1251 | All LOCK operations issued before an UNLOCK operation will be completed | |
1252 | before the UNLOCK operation. | |
1253 | ||
1254 | All UNLOCK operations issued before a LOCK operation will be completed | |
1255 | before the LOCK operation. | |
1256 | ||
1257 | (5) Failed conditional LOCK implication: | |
1258 | ||
1259 | Certain variants of the LOCK operation may fail, either due to being | |
1260 | unable to get the lock immediately, or due to receiving an unblocked | |
1261 | signal whilst asleep waiting for the lock to become available. Failed | |
1262 | locks do not imply any sort of barrier. | |
1263 | ||
1264 | Therefore, from (1), (2) and (4) an UNLOCK followed by an unconditional LOCK is | |
1265 | equivalent to a full barrier, but a LOCK followed by an UNLOCK is not. | |
1266 | ||
81fc6323 JP |
1267 | [!] Note: one of the consequences of LOCKs and UNLOCKs being only one-way |
1268 | barriers is that the effects of instructions outside of a critical section | |
1269 | may seep into the inside of the critical section. | |
108b42b4 | 1270 | |
670bd95e DH |
1271 | A LOCK followed by an UNLOCK may not be assumed to be full memory barrier |
1272 | because it is possible for an access preceding the LOCK to happen after the | |
1273 | LOCK, and an access following the UNLOCK to happen before the UNLOCK, and the | |
1274 | two accesses can themselves then cross: | |
1275 | ||
1276 | *A = a; | |
1277 | LOCK | |
1278 | UNLOCK | |
1279 | *B = b; | |
1280 | ||
1281 | may occur as: | |
1282 | ||
1283 | LOCK, STORE *B, STORE *A, UNLOCK | |
1284 | ||
108b42b4 DH |
1285 | Locks and semaphores may not provide any guarantee of ordering on UP compiled |
1286 | systems, and so cannot be counted on in such a situation to actually achieve | |
1287 | anything at all - especially with respect to I/O accesses - unless combined | |
1288 | with interrupt disabling operations. | |
1289 | ||
1290 | See also the section on "Inter-CPU locking barrier effects". | |
1291 | ||
1292 | ||
1293 | As an example, consider the following: | |
1294 | ||
1295 | *A = a; | |
1296 | *B = b; | |
1297 | LOCK | |
1298 | *C = c; | |
1299 | *D = d; | |
1300 | UNLOCK | |
1301 | *E = e; | |
1302 | *F = f; | |
1303 | ||
1304 | The following sequence of events is acceptable: | |
1305 | ||
1306 | LOCK, {*F,*A}, *E, {*C,*D}, *B, UNLOCK | |
1307 | ||
1308 | [+] Note that {*F,*A} indicates a combined access. | |
1309 | ||
1310 | But none of the following are: | |
1311 | ||
1312 | {*F,*A}, *B, LOCK, *C, *D, UNLOCK, *E | |
1313 | *A, *B, *C, LOCK, *D, UNLOCK, *E, *F | |
1314 | *A, *B, LOCK, *C, UNLOCK, *D, *E, *F | |
1315 | *B, LOCK, *C, *D, UNLOCK, {*F,*A}, *E | |
1316 | ||
1317 | ||
1318 | ||
1319 | INTERRUPT DISABLING FUNCTIONS | |
1320 | ----------------------------- | |
1321 | ||
1322 | Functions that disable interrupts (LOCK equivalent) and enable interrupts | |
1323 | (UNLOCK equivalent) will act as compiler barriers only. So if memory or I/O | |
1324 | barriers are required in such a situation, they must be provided from some | |
1325 | other means. | |
1326 | ||
1327 | ||
50fa610a DH |
1328 | SLEEP AND WAKE-UP FUNCTIONS |
1329 | --------------------------- | |
1330 | ||
1331 | Sleeping and waking on an event flagged in global data can be viewed as an | |
1332 | interaction between two pieces of data: the task state of the task waiting for | |
1333 | the event and the global data used to indicate the event. To make sure that | |
1334 | these appear to happen in the right order, the primitives to begin the process | |
1335 | of going to sleep, and the primitives to initiate a wake up imply certain | |
1336 | barriers. | |
1337 | ||
1338 | Firstly, the sleeper normally follows something like this sequence of events: | |
1339 | ||
1340 | for (;;) { | |
1341 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1342 | if (event_indicated) | |
1343 | break; | |
1344 | schedule(); | |
1345 | } | |
1346 | ||
1347 | A general memory barrier is interpolated automatically by set_current_state() | |
1348 | after it has altered the task state: | |
1349 | ||
1350 | CPU 1 | |
1351 | =============================== | |
1352 | set_current_state(); | |
1353 | set_mb(); | |
1354 | STORE current->state | |
1355 | <general barrier> | |
1356 | LOAD event_indicated | |
1357 | ||
1358 | set_current_state() may be wrapped by: | |
1359 | ||
1360 | prepare_to_wait(); | |
1361 | prepare_to_wait_exclusive(); | |
1362 | ||
1363 | which therefore also imply a general memory barrier after setting the state. | |
1364 | The whole sequence above is available in various canned forms, all of which | |
1365 | interpolate the memory barrier in the right place: | |
1366 | ||
1367 | wait_event(); | |
1368 | wait_event_interruptible(); | |
1369 | wait_event_interruptible_exclusive(); | |
1370 | wait_event_interruptible_timeout(); | |
1371 | wait_event_killable(); | |
1372 | wait_event_timeout(); | |
1373 | wait_on_bit(); | |
1374 | wait_on_bit_lock(); | |
1375 | ||
1376 | ||
1377 | Secondly, code that performs a wake up normally follows something like this: | |
1378 | ||
1379 | event_indicated = 1; | |
1380 | wake_up(&event_wait_queue); | |
1381 | ||
1382 | or: | |
1383 | ||
1384 | event_indicated = 1; | |
1385 | wake_up_process(event_daemon); | |
1386 | ||
1387 | A write memory barrier is implied by wake_up() and co. if and only if they wake | |
1388 | something up. The barrier occurs before the task state is cleared, and so sits | |
1389 | between the STORE to indicate the event and the STORE to set TASK_RUNNING: | |
1390 | ||
1391 | CPU 1 CPU 2 | |
1392 | =============================== =============================== | |
1393 | set_current_state(); STORE event_indicated | |
1394 | set_mb(); wake_up(); | |
1395 | STORE current->state <write barrier> | |
1396 | <general barrier> STORE current->state | |
1397 | LOAD event_indicated | |
1398 | ||
1399 | The available waker functions include: | |
1400 | ||
1401 | complete(); | |
1402 | wake_up(); | |
1403 | wake_up_all(); | |
1404 | wake_up_bit(); | |
1405 | wake_up_interruptible(); | |
1406 | wake_up_interruptible_all(); | |
1407 | wake_up_interruptible_nr(); | |
1408 | wake_up_interruptible_poll(); | |
1409 | wake_up_interruptible_sync(); | |
1410 | wake_up_interruptible_sync_poll(); | |
1411 | wake_up_locked(); | |
1412 | wake_up_locked_poll(); | |
1413 | wake_up_nr(); | |
1414 | wake_up_poll(); | |
1415 | wake_up_process(); | |
1416 | ||
1417 | ||
1418 | [!] Note that the memory barriers implied by the sleeper and the waker do _not_ | |
1419 | order multiple stores before the wake-up with respect to loads of those stored | |
1420 | values after the sleeper has called set_current_state(). For instance, if the | |
1421 | sleeper does: | |
1422 | ||
1423 | set_current_state(TASK_INTERRUPTIBLE); | |
1424 | if (event_indicated) | |
1425 | break; | |
1426 | __set_current_state(TASK_RUNNING); | |
1427 | do_something(my_data); | |
1428 | ||
1429 | and the waker does: | |
1430 | ||
1431 | my_data = value; | |
1432 | event_indicated = 1; | |
1433 | wake_up(&event_wait_queue); | |
1434 | ||
1435 | there's no guarantee that the change to event_indicated will be perceived by | |
1436 | the sleeper as coming after the change to my_data. In such a circumstance, the | |
1437 | code on both sides must interpolate its own memory barriers between the | |
1438 | separate data accesses. Thus the above sleeper ought to do: | |
1439 | ||
1440 | set_current_state(TASK_INTERRUPTIBLE); | |
1441 | if (event_indicated) { | |
1442 | smp_rmb(); | |
1443 | do_something(my_data); | |
1444 | } | |
1445 | ||
1446 | and the waker should do: | |
1447 | ||
1448 | my_data = value; | |
1449 | smp_wmb(); | |
1450 | event_indicated = 1; | |
1451 | wake_up(&event_wait_queue); | |
1452 | ||
1453 | ||
108b42b4 DH |
1454 | MISCELLANEOUS FUNCTIONS |
1455 | ----------------------- | |
1456 | ||
1457 | Other functions that imply barriers: | |
1458 | ||
1459 | (*) schedule() and similar imply full memory barriers. | |
1460 | ||
108b42b4 DH |
1461 | |
1462 | ================================= | |
1463 | INTER-CPU LOCKING BARRIER EFFECTS | |
1464 | ================================= | |
1465 | ||
1466 | On SMP systems locking primitives give a more substantial form of barrier: one | |
1467 | that does affect memory access ordering on other CPUs, within the context of | |
1468 | conflict on any particular lock. | |
1469 | ||
1470 | ||
1471 | LOCKS VS MEMORY ACCESSES | |
1472 | ------------------------ | |
1473 | ||
79afecfa | 1474 | Consider the following: the system has a pair of spinlocks (M) and (Q), and |
108b42b4 DH |
1475 | three CPUs; then should the following sequence of events occur: |
1476 | ||
1477 | CPU 1 CPU 2 | |
1478 | =============================== =============================== | |
2ecf8101 | 1479 | ACCESS_ONCE(*A) = a; ACCESS_ONCE(*E) = e; |
108b42b4 | 1480 | LOCK M LOCK Q |
2ecf8101 PM |
1481 | ACCESS_ONCE(*B) = b; ACCESS_ONCE(*F) = f; |
1482 | ACCESS_ONCE(*C) = c; ACCESS_ONCE(*G) = g; | |
108b42b4 | 1483 | UNLOCK M UNLOCK Q |
2ecf8101 | 1484 | ACCESS_ONCE(*D) = d; ACCESS_ONCE(*H) = h; |
108b42b4 | 1485 | |
81fc6323 | 1486 | Then there is no guarantee as to what order CPU 3 will see the accesses to *A |
108b42b4 DH |
1487 | through *H occur in, other than the constraints imposed by the separate locks |
1488 | on the separate CPUs. It might, for example, see: | |
1489 | ||
1490 | *E, LOCK M, LOCK Q, *G, *C, *F, *A, *B, UNLOCK Q, *D, *H, UNLOCK M | |
1491 | ||
1492 | But it won't see any of: | |
1493 | ||
1494 | *B, *C or *D preceding LOCK M | |
1495 | *A, *B or *C following UNLOCK M | |
1496 | *F, *G or *H preceding LOCK Q | |
1497 | *E, *F or *G following UNLOCK Q | |
1498 | ||
1499 | ||
1500 | However, if the following occurs: | |
1501 | ||
1502 | CPU 1 CPU 2 | |
1503 | =============================== =============================== | |
2ecf8101 PM |
1504 | ACCESS_ONCE(*A) = a; |
1505 | LOCK M [1] | |
1506 | ACCESS_ONCE(*B) = b; | |
1507 | ACCESS_ONCE(*C) = c; | |
1508 | UNLOCK M [1] | |
1509 | ACCESS_ONCE(*D) = d; ACCESS_ONCE(*E) = e; | |
1510 | LOCK M [2] | |
1511 | ACCESS_ONCE(*F) = f; | |
1512 | ACCESS_ONCE(*G) = g; | |
1513 | UNLOCK M [2] | |
1514 | ACCESS_ONCE(*H) = h; | |
108b42b4 | 1515 | |
81fc6323 | 1516 | CPU 3 might see: |
108b42b4 DH |
1517 | |
1518 | *E, LOCK M [1], *C, *B, *A, UNLOCK M [1], | |
1519 | LOCK M [2], *H, *F, *G, UNLOCK M [2], *D | |
1520 | ||
81fc6323 | 1521 | But assuming CPU 1 gets the lock first, CPU 3 won't see any of: |
108b42b4 DH |
1522 | |
1523 | *B, *C, *D, *F, *G or *H preceding LOCK M [1] | |
1524 | *A, *B or *C following UNLOCK M [1] | |
1525 | *F, *G or *H preceding LOCK M [2] | |
1526 | *A, *B, *C, *E, *F or *G following UNLOCK M [2] | |
1527 | ||
1528 | ||
1529 | LOCKS VS I/O ACCESSES | |
1530 | --------------------- | |
1531 | ||
1532 | Under certain circumstances (especially involving NUMA), I/O accesses within | |
1533 | two spinlocked sections on two different CPUs may be seen as interleaved by the | |
1534 | PCI bridge, because the PCI bridge does not necessarily participate in the | |
1535 | cache-coherence protocol, and is therefore incapable of issuing the required | |
1536 | read memory barriers. | |
1537 | ||
1538 | For example: | |
1539 | ||
1540 | CPU 1 CPU 2 | |
1541 | =============================== =============================== | |
1542 | spin_lock(Q) | |
1543 | writel(0, ADDR) | |
1544 | writel(1, DATA); | |
1545 | spin_unlock(Q); | |
1546 | spin_lock(Q); | |
1547 | writel(4, ADDR); | |
1548 | writel(5, DATA); | |
1549 | spin_unlock(Q); | |
1550 | ||
1551 | may be seen by the PCI bridge as follows: | |
1552 | ||
1553 | STORE *ADDR = 0, STORE *ADDR = 4, STORE *DATA = 1, STORE *DATA = 5 | |
1554 | ||
1555 | which would probably cause the hardware to malfunction. | |
1556 | ||
1557 | ||
1558 | What is necessary here is to intervene with an mmiowb() before dropping the | |
1559 | spinlock, for example: | |
1560 | ||
1561 | CPU 1 CPU 2 | |
1562 | =============================== =============================== | |
1563 | spin_lock(Q) | |
1564 | writel(0, ADDR) | |
1565 | writel(1, DATA); | |
1566 | mmiowb(); | |
1567 | spin_unlock(Q); | |
1568 | spin_lock(Q); | |
1569 | writel(4, ADDR); | |
1570 | writel(5, DATA); | |
1571 | mmiowb(); | |
1572 | spin_unlock(Q); | |
1573 | ||
81fc6323 JP |
1574 | this will ensure that the two stores issued on CPU 1 appear at the PCI bridge |
1575 | before either of the stores issued on CPU 2. | |
108b42b4 DH |
1576 | |
1577 | ||
81fc6323 JP |
1578 | Furthermore, following a store by a load from the same device obviates the need |
1579 | for the mmiowb(), because the load forces the store to complete before the load | |
108b42b4 DH |
1580 | is performed: |
1581 | ||
1582 | CPU 1 CPU 2 | |
1583 | =============================== =============================== | |
1584 | spin_lock(Q) | |
1585 | writel(0, ADDR) | |
1586 | a = readl(DATA); | |
1587 | spin_unlock(Q); | |
1588 | spin_lock(Q); | |
1589 | writel(4, ADDR); | |
1590 | b = readl(DATA); | |
1591 | spin_unlock(Q); | |
1592 | ||
1593 | ||
1594 | See Documentation/DocBook/deviceiobook.tmpl for more information. | |
1595 | ||
1596 | ||
1597 | ================================= | |
1598 | WHERE ARE MEMORY BARRIERS NEEDED? | |
1599 | ================================= | |
1600 | ||
1601 | Under normal operation, memory operation reordering is generally not going to | |
1602 | be a problem as a single-threaded linear piece of code will still appear to | |
50fa610a | 1603 | work correctly, even if it's in an SMP kernel. There are, however, four |
108b42b4 DH |
1604 | circumstances in which reordering definitely _could_ be a problem: |
1605 | ||
1606 | (*) Interprocessor interaction. | |
1607 | ||
1608 | (*) Atomic operations. | |
1609 | ||
81fc6323 | 1610 | (*) Accessing devices. |
108b42b4 DH |
1611 | |
1612 | (*) Interrupts. | |
1613 | ||
1614 | ||
1615 | INTERPROCESSOR INTERACTION | |
1616 | -------------------------- | |
1617 | ||
1618 | When there's a system with more than one processor, more than one CPU in the | |
1619 | system may be working on the same data set at the same time. This can cause | |
1620 | synchronisation problems, and the usual way of dealing with them is to use | |
1621 | locks. Locks, however, are quite expensive, and so it may be preferable to | |
1622 | operate without the use of a lock if at all possible. In such a case | |
1623 | operations that affect both CPUs may have to be carefully ordered to prevent | |
1624 | a malfunction. | |
1625 | ||
1626 | Consider, for example, the R/W semaphore slow path. Here a waiting process is | |
1627 | queued on the semaphore, by virtue of it having a piece of its stack linked to | |
1628 | the semaphore's list of waiting processes: | |
1629 | ||
1630 | struct rw_semaphore { | |
1631 | ... | |
1632 | spinlock_t lock; | |
1633 | struct list_head waiters; | |
1634 | }; | |
1635 | ||
1636 | struct rwsem_waiter { | |
1637 | struct list_head list; | |
1638 | struct task_struct *task; | |
1639 | }; | |
1640 | ||
1641 | To wake up a particular waiter, the up_read() or up_write() functions have to: | |
1642 | ||
1643 | (1) read the next pointer from this waiter's record to know as to where the | |
1644 | next waiter record is; | |
1645 | ||
81fc6323 | 1646 | (2) read the pointer to the waiter's task structure; |
108b42b4 DH |
1647 | |
1648 | (3) clear the task pointer to tell the waiter it has been given the semaphore; | |
1649 | ||
1650 | (4) call wake_up_process() on the task; and | |
1651 | ||
1652 | (5) release the reference held on the waiter's task struct. | |
1653 | ||
81fc6323 | 1654 | In other words, it has to perform this sequence of events: |
108b42b4 DH |
1655 | |
1656 | LOAD waiter->list.next; | |
1657 | LOAD waiter->task; | |
1658 | STORE waiter->task; | |
1659 | CALL wakeup | |
1660 | RELEASE task | |
1661 | ||
1662 | and if any of these steps occur out of order, then the whole thing may | |
1663 | malfunction. | |
1664 | ||
1665 | Once it has queued itself and dropped the semaphore lock, the waiter does not | |
1666 | get the lock again; it instead just waits for its task pointer to be cleared | |
1667 | before proceeding. Since the record is on the waiter's stack, this means that | |
1668 | if the task pointer is cleared _before_ the next pointer in the list is read, | |
1669 | another CPU might start processing the waiter and might clobber the waiter's | |
1670 | stack before the up*() function has a chance to read the next pointer. | |
1671 | ||
1672 | Consider then what might happen to the above sequence of events: | |
1673 | ||
1674 | CPU 1 CPU 2 | |
1675 | =============================== =============================== | |
1676 | down_xxx() | |
1677 | Queue waiter | |
1678 | Sleep | |
1679 | up_yyy() | |
1680 | LOAD waiter->task; | |
1681 | STORE waiter->task; | |
1682 | Woken up by other event | |
1683 | <preempt> | |
1684 | Resume processing | |
1685 | down_xxx() returns | |
1686 | call foo() | |
1687 | foo() clobbers *waiter | |
1688 | </preempt> | |
1689 | LOAD waiter->list.next; | |
1690 | --- OOPS --- | |
1691 | ||
1692 | This could be dealt with using the semaphore lock, but then the down_xxx() | |
1693 | function has to needlessly get the spinlock again after being woken up. | |
1694 | ||
1695 | The way to deal with this is to insert a general SMP memory barrier: | |
1696 | ||
1697 | LOAD waiter->list.next; | |
1698 | LOAD waiter->task; | |
1699 | smp_mb(); | |
1700 | STORE waiter->task; | |
1701 | CALL wakeup | |
1702 | RELEASE task | |
1703 | ||
1704 | In this case, the barrier makes a guarantee that all memory accesses before the | |
1705 | barrier will appear to happen before all the memory accesses after the barrier | |
1706 | with respect to the other CPUs on the system. It does _not_ guarantee that all | |
1707 | the memory accesses before the barrier will be complete by the time the barrier | |
1708 | instruction itself is complete. | |
1709 | ||
1710 | On a UP system - where this wouldn't be a problem - the smp_mb() is just a | |
1711 | compiler barrier, thus making sure the compiler emits the instructions in the | |
6bc39274 DH |
1712 | right order without actually intervening in the CPU. Since there's only one |
1713 | CPU, that CPU's dependency ordering logic will take care of everything else. | |
108b42b4 DH |
1714 | |
1715 | ||
1716 | ATOMIC OPERATIONS | |
1717 | ----------------- | |
1718 | ||
dbc8700e DH |
1719 | Whilst they are technically interprocessor interaction considerations, atomic |
1720 | operations are noted specially as some of them imply full memory barriers and | |
1721 | some don't, but they're very heavily relied on as a group throughout the | |
1722 | kernel. | |
1723 | ||
1724 | Any atomic operation that modifies some state in memory and returns information | |
1725 | about the state (old or new) implies an SMP-conditional general memory barrier | |
26333576 NP |
1726 | (smp_mb()) on each side of the actual operation (with the exception of |
1727 | explicit lock operations, described later). These include: | |
108b42b4 DH |
1728 | |
1729 | xchg(); | |
1730 | cmpxchg(); | |
fb2b5819 PM |
1731 | atomic_xchg(); atomic_long_xchg(); |
1732 | atomic_cmpxchg(); atomic_long_cmpxchg(); | |
1733 | atomic_inc_return(); atomic_long_inc_return(); | |
1734 | atomic_dec_return(); atomic_long_dec_return(); | |
1735 | atomic_add_return(); atomic_long_add_return(); | |
1736 | atomic_sub_return(); atomic_long_sub_return(); | |
1737 | atomic_inc_and_test(); atomic_long_inc_and_test(); | |
1738 | atomic_dec_and_test(); atomic_long_dec_and_test(); | |
1739 | atomic_sub_and_test(); atomic_long_sub_and_test(); | |
1740 | atomic_add_negative(); atomic_long_add_negative(); | |
dbc8700e DH |
1741 | test_and_set_bit(); |
1742 | test_and_clear_bit(); | |
1743 | test_and_change_bit(); | |
1744 | ||
fb2b5819 PM |
1745 | /* when succeeds (returns 1) */ |
1746 | atomic_add_unless(); atomic_long_add_unless(); | |
1747 | ||
dbc8700e DH |
1748 | These are used for such things as implementing LOCK-class and UNLOCK-class |
1749 | operations and adjusting reference counters towards object destruction, and as | |
1750 | such the implicit memory barrier effects are necessary. | |
108b42b4 | 1751 | |
108b42b4 | 1752 | |
81fc6323 | 1753 | The following operations are potential problems as they do _not_ imply memory |
dbc8700e DH |
1754 | barriers, but might be used for implementing such things as UNLOCK-class |
1755 | operations: | |
108b42b4 | 1756 | |
dbc8700e | 1757 | atomic_set(); |
108b42b4 DH |
1758 | set_bit(); |
1759 | clear_bit(); | |
1760 | change_bit(); | |
dbc8700e DH |
1761 | |
1762 | With these the appropriate explicit memory barrier should be used if necessary | |
1763 | (smp_mb__before_clear_bit() for instance). | |
108b42b4 DH |
1764 | |
1765 | ||
dbc8700e DH |
1766 | The following also do _not_ imply memory barriers, and so may require explicit |
1767 | memory barriers under some circumstances (smp_mb__before_atomic_dec() for | |
81fc6323 | 1768 | instance): |
108b42b4 DH |
1769 | |
1770 | atomic_add(); | |
1771 | atomic_sub(); | |
1772 | atomic_inc(); | |
1773 | atomic_dec(); | |
1774 | ||
1775 | If they're used for statistics generation, then they probably don't need memory | |
1776 | barriers, unless there's a coupling between statistical data. | |
1777 | ||
1778 | If they're used for reference counting on an object to control its lifetime, | |
1779 | they probably don't need memory barriers because either the reference count | |
1780 | will be adjusted inside a locked section, or the caller will already hold | |
1781 | sufficient references to make the lock, and thus a memory barrier unnecessary. | |
1782 | ||
1783 | If they're used for constructing a lock of some description, then they probably | |
1784 | do need memory barriers as a lock primitive generally has to do things in a | |
1785 | specific order. | |
1786 | ||
108b42b4 | 1787 | Basically, each usage case has to be carefully considered as to whether memory |
dbc8700e DH |
1788 | barriers are needed or not. |
1789 | ||
26333576 NP |
1790 | The following operations are special locking primitives: |
1791 | ||
1792 | test_and_set_bit_lock(); | |
1793 | clear_bit_unlock(); | |
1794 | __clear_bit_unlock(); | |
1795 | ||
1796 | These implement LOCK-class and UNLOCK-class operations. These should be used in | |
1797 | preference to other operations when implementing locking primitives, because | |
1798 | their implementations can be optimised on many architectures. | |
1799 | ||
dbc8700e DH |
1800 | [!] Note that special memory barrier primitives are available for these |
1801 | situations because on some CPUs the atomic instructions used imply full memory | |
1802 | barriers, and so barrier instructions are superfluous in conjunction with them, | |
1803 | and in such cases the special barrier primitives will be no-ops. | |
108b42b4 DH |
1804 | |
1805 | See Documentation/atomic_ops.txt for more information. | |
1806 | ||
1807 | ||
1808 | ACCESSING DEVICES | |
1809 | ----------------- | |
1810 | ||
1811 | Many devices can be memory mapped, and so appear to the CPU as if they're just | |
1812 | a set of memory locations. To control such a device, the driver usually has to | |
1813 | make the right memory accesses in exactly the right order. | |
1814 | ||
1815 | However, having a clever CPU or a clever compiler creates a potential problem | |
1816 | in that the carefully sequenced accesses in the driver code won't reach the | |
1817 | device in the requisite order if the CPU or the compiler thinks it is more | |
1818 | efficient to reorder, combine or merge accesses - something that would cause | |
1819 | the device to malfunction. | |
1820 | ||
1821 | Inside of the Linux kernel, I/O should be done through the appropriate accessor | |
1822 | routines - such as inb() or writel() - which know how to make such accesses | |
1823 | appropriately sequential. Whilst this, for the most part, renders the explicit | |
1824 | use of memory barriers unnecessary, there are a couple of situations where they | |
1825 | might be needed: | |
1826 | ||
1827 | (1) On some systems, I/O stores are not strongly ordered across all CPUs, and | |
1828 | so for _all_ general drivers locks should be used and mmiowb() must be | |
1829 | issued prior to unlocking the critical section. | |
1830 | ||
1831 | (2) If the accessor functions are used to refer to an I/O memory window with | |
1832 | relaxed memory access properties, then _mandatory_ memory barriers are | |
1833 | required to enforce ordering. | |
1834 | ||
1835 | See Documentation/DocBook/deviceiobook.tmpl for more information. | |
1836 | ||
1837 | ||
1838 | INTERRUPTS | |
1839 | ---------- | |
1840 | ||
1841 | A driver may be interrupted by its own interrupt service routine, and thus the | |
1842 | two parts of the driver may interfere with each other's attempts to control or | |
1843 | access the device. | |
1844 | ||
1845 | This may be alleviated - at least in part - by disabling local interrupts (a | |
1846 | form of locking), such that the critical operations are all contained within | |
1847 | the interrupt-disabled section in the driver. Whilst the driver's interrupt | |
1848 | routine is executing, the driver's core may not run on the same CPU, and its | |
1849 | interrupt is not permitted to happen again until the current interrupt has been | |
1850 | handled, thus the interrupt handler does not need to lock against that. | |
1851 | ||
1852 | However, consider a driver that was talking to an ethernet card that sports an | |
1853 | address register and a data register. If that driver's core talks to the card | |
1854 | under interrupt-disablement and then the driver's interrupt handler is invoked: | |
1855 | ||
1856 | LOCAL IRQ DISABLE | |
1857 | writew(ADDR, 3); | |
1858 | writew(DATA, y); | |
1859 | LOCAL IRQ ENABLE | |
1860 | <interrupt> | |
1861 | writew(ADDR, 4); | |
1862 | q = readw(DATA); | |
1863 | </interrupt> | |
1864 | ||
1865 | The store to the data register might happen after the second store to the | |
1866 | address register if ordering rules are sufficiently relaxed: | |
1867 | ||
1868 | STORE *ADDR = 3, STORE *ADDR = 4, STORE *DATA = y, q = LOAD *DATA | |
1869 | ||
1870 | ||
1871 | If ordering rules are relaxed, it must be assumed that accesses done inside an | |
1872 | interrupt disabled section may leak outside of it and may interleave with | |
1873 | accesses performed in an interrupt - and vice versa - unless implicit or | |
1874 | explicit barriers are used. | |
1875 | ||
1876 | Normally this won't be a problem because the I/O accesses done inside such | |
1877 | sections will include synchronous load operations on strictly ordered I/O | |
1878 | registers that form implicit I/O barriers. If this isn't sufficient then an | |
1879 | mmiowb() may need to be used explicitly. | |
1880 | ||
1881 | ||
1882 | A similar situation may occur between an interrupt routine and two routines | |
1883 | running on separate CPUs that communicate with each other. If such a case is | |
1884 | likely, then interrupt-disabling locks should be used to guarantee ordering. | |
1885 | ||
1886 | ||
1887 | ========================== | |
1888 | KERNEL I/O BARRIER EFFECTS | |
1889 | ========================== | |
1890 | ||
1891 | When accessing I/O memory, drivers should use the appropriate accessor | |
1892 | functions: | |
1893 | ||
1894 | (*) inX(), outX(): | |
1895 | ||
1896 | These are intended to talk to I/O space rather than memory space, but | |
1897 | that's primarily a CPU-specific concept. The i386 and x86_64 processors do | |
1898 | indeed have special I/O space access cycles and instructions, but many | |
1899 | CPUs don't have such a concept. | |
1900 | ||
81fc6323 JP |
1901 | The PCI bus, amongst others, defines an I/O space concept which - on such |
1902 | CPUs as i386 and x86_64 - readily maps to the CPU's concept of I/O | |
6bc39274 DH |
1903 | space. However, it may also be mapped as a virtual I/O space in the CPU's |
1904 | memory map, particularly on those CPUs that don't support alternate I/O | |
1905 | spaces. | |
108b42b4 DH |
1906 | |
1907 | Accesses to this space may be fully synchronous (as on i386), but | |
1908 | intermediary bridges (such as the PCI host bridge) may not fully honour | |
1909 | that. | |
1910 | ||
1911 | They are guaranteed to be fully ordered with respect to each other. | |
1912 | ||
1913 | They are not guaranteed to be fully ordered with respect to other types of | |
1914 | memory and I/O operation. | |
1915 | ||
1916 | (*) readX(), writeX(): | |
1917 | ||
1918 | Whether these are guaranteed to be fully ordered and uncombined with | |
1919 | respect to each other on the issuing CPU depends on the characteristics | |
1920 | defined for the memory window through which they're accessing. On later | |
1921 | i386 architecture machines, for example, this is controlled by way of the | |
1922 | MTRR registers. | |
1923 | ||
81fc6323 | 1924 | Ordinarily, these will be guaranteed to be fully ordered and uncombined, |
108b42b4 DH |
1925 | provided they're not accessing a prefetchable device. |
1926 | ||
1927 | However, intermediary hardware (such as a PCI bridge) may indulge in | |
1928 | deferral if it so wishes; to flush a store, a load from the same location | |
1929 | is preferred[*], but a load from the same device or from configuration | |
1930 | space should suffice for PCI. | |
1931 | ||
1932 | [*] NOTE! attempting to load from the same location as was written to may | |
e0edc78f IM |
1933 | cause a malfunction - consider the 16550 Rx/Tx serial registers for |
1934 | example. | |
108b42b4 DH |
1935 | |
1936 | Used with prefetchable I/O memory, an mmiowb() barrier may be required to | |
1937 | force stores to be ordered. | |
1938 | ||
1939 | Please refer to the PCI specification for more information on interactions | |
1940 | between PCI transactions. | |
1941 | ||
1942 | (*) readX_relaxed() | |
1943 | ||
1944 | These are similar to readX(), but are not guaranteed to be ordered in any | |
1945 | way. Be aware that there is no I/O read barrier available. | |
1946 | ||
1947 | (*) ioreadX(), iowriteX() | |
1948 | ||
81fc6323 | 1949 | These will perform appropriately for the type of access they're actually |
108b42b4 DH |
1950 | doing, be it inX()/outX() or readX()/writeX(). |
1951 | ||
1952 | ||
1953 | ======================================== | |
1954 | ASSUMED MINIMUM EXECUTION ORDERING MODEL | |
1955 | ======================================== | |
1956 | ||
1957 | It has to be assumed that the conceptual CPU is weakly-ordered but that it will | |
1958 | maintain the appearance of program causality with respect to itself. Some CPUs | |
1959 | (such as i386 or x86_64) are more constrained than others (such as powerpc or | |
1960 | frv), and so the most relaxed case (namely DEC Alpha) must be assumed outside | |
1961 | of arch-specific code. | |
1962 | ||
1963 | This means that it must be considered that the CPU will execute its instruction | |
1964 | stream in any order it feels like - or even in parallel - provided that if an | |
81fc6323 | 1965 | instruction in the stream depends on an earlier instruction, then that |
108b42b4 DH |
1966 | earlier instruction must be sufficiently complete[*] before the later |
1967 | instruction may proceed; in other words: provided that the appearance of | |
1968 | causality is maintained. | |
1969 | ||
1970 | [*] Some instructions have more than one effect - such as changing the | |
1971 | condition codes, changing registers or changing memory - and different | |
1972 | instructions may depend on different effects. | |
1973 | ||
1974 | A CPU may also discard any instruction sequence that winds up having no | |
1975 | ultimate effect. For example, if two adjacent instructions both load an | |
1976 | immediate value into the same register, the first may be discarded. | |
1977 | ||
1978 | ||
1979 | Similarly, it has to be assumed that compiler might reorder the instruction | |
1980 | stream in any way it sees fit, again provided the appearance of causality is | |
1981 | maintained. | |
1982 | ||
1983 | ||
1984 | ============================ | |
1985 | THE EFFECTS OF THE CPU CACHE | |
1986 | ============================ | |
1987 | ||
1988 | The way cached memory operations are perceived across the system is affected to | |
1989 | a certain extent by the caches that lie between CPUs and memory, and by the | |
1990 | memory coherence system that maintains the consistency of state in the system. | |
1991 | ||
1992 | As far as the way a CPU interacts with another part of the system through the | |
1993 | caches goes, the memory system has to include the CPU's caches, and memory | |
1994 | barriers for the most part act at the interface between the CPU and its cache | |
1995 | (memory barriers logically act on the dotted line in the following diagram): | |
1996 | ||
1997 | <--- CPU ---> : <----------- Memory -----------> | |
1998 | : | |
1999 | +--------+ +--------+ : +--------+ +-----------+ | |
2000 | | | | | : | | | | +--------+ | |
e0edc78f IM |
2001 | | CPU | | Memory | : | CPU | | | | | |
2002 | | Core |--->| Access |----->| Cache |<-->| | | | | |
108b42b4 | 2003 | | | | Queue | : | | | |--->| Memory | |
e0edc78f IM |
2004 | | | | | : | | | | | | |
2005 | +--------+ +--------+ : +--------+ | | | | | |
108b42b4 DH |
2006 | : | Cache | +--------+ |
2007 | : | Coherency | | |
2008 | : | Mechanism | +--------+ | |
2009 | +--------+ +--------+ : +--------+ | | | | | |
2010 | | | | | : | | | | | | | |
2011 | | CPU | | Memory | : | CPU | | |--->| Device | | |
e0edc78f IM |
2012 | | Core |--->| Access |----->| Cache |<-->| | | | |
2013 | | | | Queue | : | | | | | | | |
108b42b4 DH |
2014 | | | | | : | | | | +--------+ |
2015 | +--------+ +--------+ : +--------+ +-----------+ | |
2016 | : | |
2017 | : | |
2018 | ||
2019 | Although any particular load or store may not actually appear outside of the | |
2020 | CPU that issued it since it may have been satisfied within the CPU's own cache, | |
2021 | it will still appear as if the full memory access had taken place as far as the | |
2022 | other CPUs are concerned since the cache coherency mechanisms will migrate the | |
2023 | cacheline over to the accessing CPU and propagate the effects upon conflict. | |
2024 | ||
2025 | The CPU core may execute instructions in any order it deems fit, provided the | |
2026 | expected program causality appears to be maintained. Some of the instructions | |
2027 | generate load and store operations which then go into the queue of memory | |
2028 | accesses to be performed. The core may place these in the queue in any order | |
2029 | it wishes, and continue execution until it is forced to wait for an instruction | |
2030 | to complete. | |
2031 | ||
2032 | What memory barriers are concerned with is controlling the order in which | |
2033 | accesses cross from the CPU side of things to the memory side of things, and | |
2034 | the order in which the effects are perceived to happen by the other observers | |
2035 | in the system. | |
2036 | ||
2037 | [!] Memory barriers are _not_ needed within a given CPU, as CPUs always see | |
2038 | their own loads and stores as if they had happened in program order. | |
2039 | ||
2040 | [!] MMIO or other device accesses may bypass the cache system. This depends on | |
2041 | the properties of the memory window through which devices are accessed and/or | |
2042 | the use of any special device communication instructions the CPU may have. | |
2043 | ||
2044 | ||
2045 | CACHE COHERENCY | |
2046 | --------------- | |
2047 | ||
2048 | Life isn't quite as simple as it may appear above, however: for while the | |
2049 | caches are expected to be coherent, there's no guarantee that that coherency | |
2050 | will be ordered. This means that whilst changes made on one CPU will | |
2051 | eventually become visible on all CPUs, there's no guarantee that they will | |
2052 | become apparent in the same order on those other CPUs. | |
2053 | ||
2054 | ||
81fc6323 JP |
2055 | Consider dealing with a system that has a pair of CPUs (1 & 2), each of which |
2056 | has a pair of parallel data caches (CPU 1 has A/B, and CPU 2 has C/D): | |
108b42b4 DH |
2057 | |
2058 | : | |
2059 | : +--------+ | |
2060 | : +---------+ | | | |
2061 | +--------+ : +--->| Cache A |<------->| | | |
2062 | | | : | +---------+ | | | |
2063 | | CPU 1 |<---+ | | | |
2064 | | | : | +---------+ | | | |
2065 | +--------+ : +--->| Cache B |<------->| | | |
2066 | : +---------+ | | | |
2067 | : | Memory | | |
2068 | : +---------+ | System | | |
2069 | +--------+ : +--->| Cache C |<------->| | | |
2070 | | | : | +---------+ | | | |
2071 | | CPU 2 |<---+ | | | |
2072 | | | : | +---------+ | | | |
2073 | +--------+ : +--->| Cache D |<------->| | | |
2074 | : +---------+ | | | |
2075 | : +--------+ | |
2076 | : | |
2077 | ||
2078 | Imagine the system has the following properties: | |
2079 | ||
2080 | (*) an odd-numbered cache line may be in cache A, cache C or it may still be | |
2081 | resident in memory; | |
2082 | ||
2083 | (*) an even-numbered cache line may be in cache B, cache D or it may still be | |
2084 | resident in memory; | |
2085 | ||
2086 | (*) whilst the CPU core is interrogating one cache, the other cache may be | |
2087 | making use of the bus to access the rest of the system - perhaps to | |
2088 | displace a dirty cacheline or to do a speculative load; | |
2089 | ||
2090 | (*) each cache has a queue of operations that need to be applied to that cache | |
2091 | to maintain coherency with the rest of the system; | |
2092 | ||
2093 | (*) the coherency queue is not flushed by normal loads to lines already | |
2094 | present in the cache, even though the contents of the queue may | |
81fc6323 | 2095 | potentially affect those loads. |
108b42b4 DH |
2096 | |
2097 | Imagine, then, that two writes are made on the first CPU, with a write barrier | |
2098 | between them to guarantee that they will appear to reach that CPU's caches in | |
2099 | the requisite order: | |
2100 | ||
2101 | CPU 1 CPU 2 COMMENT | |
2102 | =============== =============== ======================================= | |
2103 | u == 0, v == 1 and p == &u, q == &u | |
2104 | v = 2; | |
81fc6323 | 2105 | smp_wmb(); Make sure change to v is visible before |
108b42b4 DH |
2106 | change to p |
2107 | <A:modify v=2> v is now in cache A exclusively | |
2108 | p = &v; | |
2109 | <B:modify p=&v> p is now in cache B exclusively | |
2110 | ||
2111 | The write memory barrier forces the other CPUs in the system to perceive that | |
2112 | the local CPU's caches have apparently been updated in the correct order. But | |
81fc6323 | 2113 | now imagine that the second CPU wants to read those values: |
108b42b4 DH |
2114 | |
2115 | CPU 1 CPU 2 COMMENT | |
2116 | =============== =============== ======================================= | |
2117 | ... | |
2118 | q = p; | |
2119 | x = *q; | |
2120 | ||
81fc6323 | 2121 | The above pair of reads may then fail to happen in the expected order, as the |
108b42b4 DH |
2122 | cacheline holding p may get updated in one of the second CPU's caches whilst |
2123 | the update to the cacheline holding v is delayed in the other of the second | |
2124 | CPU's caches by some other cache event: | |
2125 | ||
2126 | CPU 1 CPU 2 COMMENT | |
2127 | =============== =============== ======================================= | |
2128 | u == 0, v == 1 and p == &u, q == &u | |
2129 | v = 2; | |
2130 | smp_wmb(); | |
2131 | <A:modify v=2> <C:busy> | |
2132 | <C:queue v=2> | |
79afecfa | 2133 | p = &v; q = p; |
108b42b4 DH |
2134 | <D:request p> |
2135 | <B:modify p=&v> <D:commit p=&v> | |
e0edc78f | 2136 | <D:read p> |
108b42b4 DH |
2137 | x = *q; |
2138 | <C:read *q> Reads from v before v updated in cache | |
2139 | <C:unbusy> | |
2140 | <C:commit v=2> | |
2141 | ||
2142 | Basically, whilst both cachelines will be updated on CPU 2 eventually, there's | |
2143 | no guarantee that, without intervention, the order of update will be the same | |
2144 | as that committed on CPU 1. | |
2145 | ||
2146 | ||
2147 | To intervene, we need to interpolate a data dependency barrier or a read | |
2148 | barrier between the loads. This will force the cache to commit its coherency | |
2149 | queue before processing any further requests: | |
2150 | ||
2151 | CPU 1 CPU 2 COMMENT | |
2152 | =============== =============== ======================================= | |
2153 | u == 0, v == 1 and p == &u, q == &u | |
2154 | v = 2; | |
2155 | smp_wmb(); | |
2156 | <A:modify v=2> <C:busy> | |
2157 | <C:queue v=2> | |
3fda982c | 2158 | p = &v; q = p; |
108b42b4 DH |
2159 | <D:request p> |
2160 | <B:modify p=&v> <D:commit p=&v> | |
e0edc78f | 2161 | <D:read p> |
108b42b4 DH |
2162 | smp_read_barrier_depends() |
2163 | <C:unbusy> | |
2164 | <C:commit v=2> | |
2165 | x = *q; | |
2166 | <C:read *q> Reads from v after v updated in cache | |
2167 | ||
2168 | ||
2169 | This sort of problem can be encountered on DEC Alpha processors as they have a | |
2170 | split cache that improves performance by making better use of the data bus. | |
2171 | Whilst most CPUs do imply a data dependency barrier on the read when a memory | |
2172 | access depends on a read, not all do, so it may not be relied on. | |
2173 | ||
2174 | Other CPUs may also have split caches, but must coordinate between the various | |
3f6dee9b | 2175 | cachelets for normal memory accesses. The semantics of the Alpha removes the |
81fc6323 | 2176 | need for coordination in the absence of memory barriers. |
108b42b4 DH |
2177 | |
2178 | ||
2179 | CACHE COHERENCY VS DMA | |
2180 | ---------------------- | |
2181 | ||
2182 | Not all systems maintain cache coherency with respect to devices doing DMA. In | |
2183 | such cases, a device attempting DMA may obtain stale data from RAM because | |
2184 | dirty cache lines may be resident in the caches of various CPUs, and may not | |
2185 | have been written back to RAM yet. To deal with this, the appropriate part of | |
2186 | the kernel must flush the overlapping bits of cache on each CPU (and maybe | |
2187 | invalidate them as well). | |
2188 | ||
2189 | In addition, the data DMA'd to RAM by a device may be overwritten by dirty | |
2190 | cache lines being written back to RAM from a CPU's cache after the device has | |
81fc6323 JP |
2191 | installed its own data, or cache lines present in the CPU's cache may simply |
2192 | obscure the fact that RAM has been updated, until at such time as the cacheline | |
2193 | is discarded from the CPU's cache and reloaded. To deal with this, the | |
2194 | appropriate part of the kernel must invalidate the overlapping bits of the | |
108b42b4 DH |
2195 | cache on each CPU. |
2196 | ||
2197 | See Documentation/cachetlb.txt for more information on cache management. | |
2198 | ||
2199 | ||
2200 | CACHE COHERENCY VS MMIO | |
2201 | ----------------------- | |
2202 | ||
2203 | Memory mapped I/O usually takes place through memory locations that are part of | |
81fc6323 | 2204 | a window in the CPU's memory space that has different properties assigned than |
108b42b4 DH |
2205 | the usual RAM directed window. |
2206 | ||
2207 | Amongst these properties is usually the fact that such accesses bypass the | |
2208 | caching entirely and go directly to the device buses. This means MMIO accesses | |
2209 | may, in effect, overtake accesses to cached memory that were emitted earlier. | |
2210 | A memory barrier isn't sufficient in such a case, but rather the cache must be | |
2211 | flushed between the cached memory write and the MMIO access if the two are in | |
2212 | any way dependent. | |
2213 | ||
2214 | ||
2215 | ========================= | |
2216 | THE THINGS CPUS GET UP TO | |
2217 | ========================= | |
2218 | ||
2219 | A programmer might take it for granted that the CPU will perform memory | |
81fc6323 | 2220 | operations in exactly the order specified, so that if the CPU is, for example, |
108b42b4 DH |
2221 | given the following piece of code to execute: |
2222 | ||
2ecf8101 PM |
2223 | a = ACCESS_ONCE(*A); |
2224 | ACCESS_ONCE(*B) = b; | |
2225 | c = ACCESS_ONCE(*C); | |
2226 | d = ACCESS_ONCE(*D); | |
2227 | ACCESS_ONCE(*E) = e; | |
108b42b4 | 2228 | |
81fc6323 | 2229 | they would then expect that the CPU will complete the memory operation for each |
108b42b4 DH |
2230 | instruction before moving on to the next one, leading to a definite sequence of |
2231 | operations as seen by external observers in the system: | |
2232 | ||
2233 | LOAD *A, STORE *B, LOAD *C, LOAD *D, STORE *E. | |
2234 | ||
2235 | ||
2236 | Reality is, of course, much messier. With many CPUs and compilers, the above | |
2237 | assumption doesn't hold because: | |
2238 | ||
2239 | (*) loads are more likely to need to be completed immediately to permit | |
2240 | execution progress, whereas stores can often be deferred without a | |
2241 | problem; | |
2242 | ||
2243 | (*) loads may be done speculatively, and the result discarded should it prove | |
2244 | to have been unnecessary; | |
2245 | ||
81fc6323 JP |
2246 | (*) loads may be done speculatively, leading to the result having been fetched |
2247 | at the wrong time in the expected sequence of events; | |
108b42b4 DH |
2248 | |
2249 | (*) the order of the memory accesses may be rearranged to promote better use | |
2250 | of the CPU buses and caches; | |
2251 | ||
2252 | (*) loads and stores may be combined to improve performance when talking to | |
2253 | memory or I/O hardware that can do batched accesses of adjacent locations, | |
2254 | thus cutting down on transaction setup costs (memory and PCI devices may | |
2255 | both be able to do this); and | |
2256 | ||
2257 | (*) the CPU's data cache may affect the ordering, and whilst cache-coherency | |
2258 | mechanisms may alleviate this - once the store has actually hit the cache | |
2259 | - there's no guarantee that the coherency management will be propagated in | |
2260 | order to other CPUs. | |
2261 | ||
2262 | So what another CPU, say, might actually observe from the above piece of code | |
2263 | is: | |
2264 | ||
2265 | LOAD *A, ..., LOAD {*C,*D}, STORE *E, STORE *B | |
2266 | ||
2267 | (Where "LOAD {*C,*D}" is a combined load) | |
2268 | ||
2269 | ||
2270 | However, it is guaranteed that a CPU will be self-consistent: it will see its | |
2271 | _own_ accesses appear to be correctly ordered, without the need for a memory | |
2272 | barrier. For instance with the following code: | |
2273 | ||
2ecf8101 PM |
2274 | U = ACCESS_ONCE(*A); |
2275 | ACCESS_ONCE(*A) = V; | |
2276 | ACCESS_ONCE(*A) = W; | |
2277 | X = ACCESS_ONCE(*A); | |
2278 | ACCESS_ONCE(*A) = Y; | |
2279 | Z = ACCESS_ONCE(*A); | |
108b42b4 DH |
2280 | |
2281 | and assuming no intervention by an external influence, it can be assumed that | |
2282 | the final result will appear to be: | |
2283 | ||
2284 | U == the original value of *A | |
2285 | X == W | |
2286 | Z == Y | |
2287 | *A == Y | |
2288 | ||
2289 | The code above may cause the CPU to generate the full sequence of memory | |
2290 | accesses: | |
2291 | ||
2292 | U=LOAD *A, STORE *A=V, STORE *A=W, X=LOAD *A, STORE *A=Y, Z=LOAD *A | |
2293 | ||
2294 | in that order, but, without intervention, the sequence may have almost any | |
2295 | combination of elements combined or discarded, provided the program's view of | |
2ecf8101 PM |
2296 | the world remains consistent. Note that ACCESS_ONCE() is -not- optional |
2297 | in the above example, as there are architectures where a given CPU might | |
2298 | interchange successive loads to the same location. On such architectures, | |
2299 | ACCESS_ONCE() does whatever is necessary to prevent this, for example, on | |
2300 | Itanium the volatile casts used by ACCESS_ONCE() cause GCC to emit the | |
2301 | special ld.acq and st.rel instructions that prevent such reordering. | |
108b42b4 DH |
2302 | |
2303 | The compiler may also combine, discard or defer elements of the sequence before | |
2304 | the CPU even sees them. | |
2305 | ||
2306 | For instance: | |
2307 | ||
2308 | *A = V; | |
2309 | *A = W; | |
2310 | ||
2311 | may be reduced to: | |
2312 | ||
2313 | *A = W; | |
2314 | ||
2ecf8101 PM |
2315 | since, without either a write barrier or an ACCESS_ONCE(), it can be |
2316 | assumed that the effect of the storage of V to *A is lost. Similarly: | |
108b42b4 DH |
2317 | |
2318 | *A = Y; | |
2319 | Z = *A; | |
2320 | ||
2ecf8101 | 2321 | may, without a memory barrier or an ACCESS_ONCE(), be reduced to: |
108b42b4 DH |
2322 | |
2323 | *A = Y; | |
2324 | Z = Y; | |
2325 | ||
2326 | and the LOAD operation never appear outside of the CPU. | |
2327 | ||
2328 | ||
2329 | AND THEN THERE'S THE ALPHA | |
2330 | -------------------------- | |
2331 | ||
2332 | The DEC Alpha CPU is one of the most relaxed CPUs there is. Not only that, | |
2333 | some versions of the Alpha CPU have a split data cache, permitting them to have | |
81fc6323 | 2334 | two semantically-related cache lines updated at separate times. This is where |
108b42b4 DH |
2335 | the data dependency barrier really becomes necessary as this synchronises both |
2336 | caches with the memory coherence system, thus making it seem like pointer | |
2337 | changes vs new data occur in the right order. | |
2338 | ||
81fc6323 | 2339 | The Alpha defines the Linux kernel's memory barrier model. |
108b42b4 DH |
2340 | |
2341 | See the subsection on "Cache Coherency" above. | |
2342 | ||
2343 | ||
90fddabf DH |
2344 | ============ |
2345 | EXAMPLE USES | |
2346 | ============ | |
2347 | ||
2348 | CIRCULAR BUFFERS | |
2349 | ---------------- | |
2350 | ||
2351 | Memory barriers can be used to implement circular buffering without the need | |
2352 | of a lock to serialise the producer with the consumer. See: | |
2353 | ||
2354 | Documentation/circular-buffers.txt | |
2355 | ||
2356 | for details. | |
2357 | ||
2358 | ||
108b42b4 DH |
2359 | ========== |
2360 | REFERENCES | |
2361 | ========== | |
2362 | ||
2363 | Alpha AXP Architecture Reference Manual, Second Edition (Sites & Witek, | |
2364 | Digital Press) | |
2365 | Chapter 5.2: Physical Address Space Characteristics | |
2366 | Chapter 5.4: Caches and Write Buffers | |
2367 | Chapter 5.5: Data Sharing | |
2368 | Chapter 5.6: Read/Write Ordering | |
2369 | ||
2370 | AMD64 Architecture Programmer's Manual Volume 2: System Programming | |
2371 | Chapter 7.1: Memory-Access Ordering | |
2372 | Chapter 7.4: Buffering and Combining Memory Writes | |
2373 | ||
2374 | IA-32 Intel Architecture Software Developer's Manual, Volume 3: | |
2375 | System Programming Guide | |
2376 | Chapter 7.1: Locked Atomic Operations | |
2377 | Chapter 7.2: Memory Ordering | |
2378 | Chapter 7.4: Serializing Instructions | |
2379 | ||
2380 | The SPARC Architecture Manual, Version 9 | |
2381 | Chapter 8: Memory Models | |
2382 | Appendix D: Formal Specification of the Memory Models | |
2383 | Appendix J: Programming with the Memory Models | |
2384 | ||
2385 | UltraSPARC Programmer Reference Manual | |
2386 | Chapter 5: Memory Accesses and Cacheability | |
2387 | Chapter 15: Sparc-V9 Memory Models | |
2388 | ||
2389 | UltraSPARC III Cu User's Manual | |
2390 | Chapter 9: Memory Models | |
2391 | ||
2392 | UltraSPARC IIIi Processor User's Manual | |
2393 | Chapter 8: Memory Models | |
2394 | ||
2395 | UltraSPARC Architecture 2005 | |
2396 | Chapter 9: Memory | |
2397 | Appendix D: Formal Specifications of the Memory Models | |
2398 | ||
2399 | UltraSPARC T1 Supplement to the UltraSPARC Architecture 2005 | |
2400 | Chapter 8: Memory Models | |
2401 | Appendix F: Caches and Cache Coherency | |
2402 | ||
2403 | Solaris Internals, Core Kernel Architecture, p63-68: | |
2404 | Chapter 3.3: Hardware Considerations for Locks and | |
2405 | Synchronization | |
2406 | ||
2407 | Unix Systems for Modern Architectures, Symmetric Multiprocessing and Caching | |
2408 | for Kernel Programmers: | |
2409 | Chapter 13: Other Memory Models | |
2410 | ||
2411 | Intel Itanium Architecture Software Developer's Manual: Volume 1: | |
2412 | Section 2.6: Speculation | |
2413 | Section 4.4: Memory Access |