blob: 650657c5473340dcd8f5ffe5f097c6776a754441 [file] [log] [blame]
David Howells108b42b2006-03-31 16:00:29 +01001 ============================
2 LINUX KERNEL MEMORY BARRIERS
3 ============================
4
5By: David Howells <dhowells@redhat.com>
6
7Contents:
8
9 (*) Abstract memory access model.
10
11 - Device operations.
12 - Guarantees.
13
14 (*) What are memory barriers?
15
16 - Varieties of memory barrier.
17 - What may not be assumed about memory barriers?
18 - Data dependency barriers.
19 - Control dependencies.
20 - SMP barrier pairing.
21 - Examples of memory barrier sequences.
David Howells670bd952006-06-10 09:54:12 -070022 - Read memory barriers vs load speculation.
David Howells108b42b2006-03-31 16:00:29 +010023
24 (*) Explicit kernel barriers.
25
26 - Compiler barrier.
Jarek Poplawski81fc6322007-05-23 13:58:20 -070027 - CPU memory barriers.
David Howells108b42b2006-03-31 16:00:29 +010028 - MMIO write barrier.
29
30 (*) Implicit kernel memory barriers.
31
32 - Locking functions.
33 - Interrupt disabling functions.
34 - Miscellaneous functions.
35
36 (*) Inter-CPU locking barrier effects.
37
38 - Locks vs memory accesses.
39 - Locks vs I/O accesses.
40
41 (*) Where are memory barriers needed?
42
43 - Interprocessor interaction.
44 - Atomic operations.
45 - Accessing devices.
46 - Interrupts.
47
48 (*) Kernel I/O barrier effects.
49
50 (*) Assumed minimum execution ordering model.
51
52 (*) The effects of the cpu cache.
53
54 - Cache coherency.
55 - Cache coherency vs DMA.
56 - Cache coherency vs MMIO.
57
58 (*) The things CPUs get up to.
59
60 - And then there's the Alpha.
61
62 (*) References.
63
64
65============================
66ABSTRACT MEMORY ACCESS MODEL
67============================
68
69Consider the following abstract model of the system:
70
71 : :
72 : :
73 : :
74 +-------+ : +--------+ : +-------+
75 | | : | | : | |
76 | | : | | : | |
77 | CPU 1 |<----->| Memory |<----->| CPU 2 |
78 | | : | | : | |
79 | | : | | : | |
80 +-------+ : +--------+ : +-------+
81 ^ : ^ : ^
82 | : | : |
83 | : | : |
84 | : v : |
85 | : +--------+ : |
86 | : | | : |
87 | : | | : |
88 +---------->| Device |<----------+
89 : | | :
90 : | | :
91 : +--------+ :
92 : :
93
94Each CPU executes a program that generates memory access operations. In the
95abstract CPU, memory operation ordering is very relaxed, and a CPU may actually
96perform the memory operations in any order it likes, provided program causality
97appears to be maintained. Similarly, the compiler may also arrange the
98instructions it emits in any order it likes, provided it doesn't affect the
99apparent operation of the program.
100
101So in the above diagram, the effects of the memory operations performed by a
102CPU are perceived by the rest of the system as the operations cross the
103interface between the CPU and rest of the system (the dotted lines).
104
105
106For example, consider the following sequence of events:
107
108 CPU 1 CPU 2
109 =============== ===============
110 { A == 1; B == 2 }
111 A = 3; x = A;
112 B = 4; y = B;
113
114The set of accesses as seen by the memory system in the middle can be arranged
115in 24 different combinations:
116
117 STORE A=3, STORE B=4, x=LOAD A->3, y=LOAD B->4
118 STORE A=3, STORE B=4, y=LOAD B->4, x=LOAD A->3
119 STORE A=3, x=LOAD A->3, STORE B=4, y=LOAD B->4
120 STORE A=3, x=LOAD A->3, y=LOAD B->2, STORE B=4
121 STORE A=3, y=LOAD B->2, STORE B=4, x=LOAD A->3
122 STORE A=3, y=LOAD B->2, x=LOAD A->3, STORE B=4
123 STORE B=4, STORE A=3, x=LOAD A->3, y=LOAD B->4
124 STORE B=4, ...
125 ...
126
127and can thus result in four different combinations of values:
128
129 x == 1, y == 2
130 x == 1, y == 4
131 x == 3, y == 2
132 x == 3, y == 4
133
134
135Furthermore, the stores committed by a CPU to the memory system may not be
136perceived by the loads made by another CPU in the same order as the stores were
137committed.
138
139
140As a further example, consider this sequence of events:
141
142 CPU 1 CPU 2
143 =============== ===============
144 { A == 1, B == 2, C = 3, P == &A, Q == &C }
145 B = 4; Q = P;
146 P = &B D = *Q;
147
148There is an obvious data dependency here, as the value loaded into D depends on
149the address retrieved from P by CPU 2. At the end of the sequence, any of the
150following results are possible:
151
152 (Q == &A) and (D == 1)
153 (Q == &B) and (D == 2)
154 (Q == &B) and (D == 4)
155
156Note that CPU 2 will never try and load C into D because the CPU will load P
157into Q before issuing the load of *Q.
158
159
160DEVICE OPERATIONS
161-----------------
162
163Some devices present their control interfaces as collections of memory
164locations, but the order in which the control registers are accessed is very
165important. For instance, imagine an ethernet card with a set of internal
166registers that are accessed through an address port register (A) and a data
167port register (D). To read internal register 5, the following code might then
168be used:
169
170 *A = 5;
171 x = *D;
172
173but this might show up as either of the following two sequences:
174
175 STORE *A = 5, x = LOAD *D
176 x = LOAD *D, STORE *A = 5
177
178the second of which will almost certainly result in a malfunction, since it set
179the address _after_ attempting to read the register.
180
181
182GUARANTEES
183----------
184
185There are some minimal guarantees that may be expected of a CPU:
186
187 (*) On any given CPU, dependent memory accesses will be issued in order, with
188 respect to itself. This means that for:
189
190 Q = P; D = *Q;
191
192 the CPU will issue the following memory operations:
193
194 Q = LOAD P, D = LOAD *Q
195
196 and always in that order.
197
198 (*) Overlapping loads and stores within a particular CPU will appear to be
199 ordered within that CPU. This means that for:
200
201 a = *X; *X = b;
202
203 the CPU will only issue the following sequence of memory operations:
204
205 a = LOAD *X, STORE *X = b
206
207 And for:
208
209 *X = c; d = *X;
210
211 the CPU will only issue:
212
213 STORE *X = c, d = LOAD *X
214
Matt LaPlantefa00e7e2006-11-30 04:55:36 +0100215 (Loads and stores overlap if they are targeted at overlapping pieces of
David Howells108b42b2006-03-31 16:00:29 +0100216 memory).
217
218And there are a number of things that _must_ or _must_not_ be assumed:
219
220 (*) It _must_not_ be assumed that independent loads and stores will be issued
221 in the order given. This means that for:
222
223 X = *A; Y = *B; *D = Z;
224
225 we may get any of the following sequences:
226
227 X = LOAD *A, Y = LOAD *B, STORE *D = Z
228 X = LOAD *A, STORE *D = Z, Y = LOAD *B
229 Y = LOAD *B, X = LOAD *A, STORE *D = Z
230 Y = LOAD *B, STORE *D = Z, X = LOAD *A
231 STORE *D = Z, X = LOAD *A, Y = LOAD *B
232 STORE *D = Z, Y = LOAD *B, X = LOAD *A
233
234 (*) It _must_ be assumed that overlapping memory accesses may be merged or
235 discarded. This means that for:
236
237 X = *A; Y = *(A + 4);
238
239 we may get any one of the following sequences:
240
241 X = LOAD *A; Y = LOAD *(A + 4);
242 Y = LOAD *(A + 4); X = LOAD *A;
243 {X, Y} = LOAD {*A, *(A + 4) };
244
245 And for:
246
247 *A = X; Y = *A;
248
249 we may get either of:
250
251 STORE *A = X; Y = LOAD *A;
David Howells670bd952006-06-10 09:54:12 -0700252 STORE *A = Y = X;
David Howells108b42b2006-03-31 16:00:29 +0100253
254
255=========================
256WHAT ARE MEMORY BARRIERS?
257=========================
258
259As can be seen above, independent memory operations are effectively performed
260in random order, but this can be a problem for CPU-CPU interaction and for I/O.
261What is required is some way of intervening to instruct the compiler and the
262CPU to restrict the order.
263
264Memory barriers are such interventions. They impose a perceived partial
David Howells2b948952006-06-25 05:48:49 -0700265ordering over the memory operations on either side of the barrier.
266
267Such enforcement is important because the CPUs and other devices in a system
Jarek Poplawski81fc6322007-05-23 13:58:20 -0700268can use a variety of tricks to improve performance, including reordering,
David Howells2b948952006-06-25 05:48:49 -0700269deferral and combination of memory operations; speculative loads; speculative
270branch prediction and various types of caching. Memory barriers are used to
271override or suppress these tricks, allowing the code to sanely control the
272interaction of multiple CPUs and/or devices.
David Howells108b42b2006-03-31 16:00:29 +0100273
274
275VARIETIES OF MEMORY BARRIER
276---------------------------
277
278Memory barriers come in four basic varieties:
279
280 (1) Write (or store) memory barriers.
281
282 A write memory barrier gives a guarantee that all the STORE operations
283 specified before the barrier will appear to happen before all the STORE
284 operations specified after the barrier with respect to the other
285 components of the system.
286
287 A write barrier is a partial ordering on stores only; it is not required
288 to have any effect on loads.
289
David Howells6bc39272006-06-25 05:49:22 -0700290 A CPU can be viewed as committing a sequence of store operations to the
David Howells108b42b2006-03-31 16:00:29 +0100291 memory system as time progresses. All stores before a write barrier will
292 occur in the sequence _before_ all the stores after the write barrier.
293
294 [!] Note that write barriers should normally be paired with read or data
295 dependency barriers; see the "SMP barrier pairing" subsection.
296
297
298 (2) Data dependency barriers.
299
300 A data dependency barrier is a weaker form of read barrier. In the case
301 where two loads are performed such that the second depends on the result
302 of the first (eg: the first load retrieves the address to which the second
303 load will be directed), a data dependency barrier would be required to
304 make sure that the target of the second load is updated before the address
305 obtained by the first load is accessed.
306
307 A data dependency barrier is a partial ordering on interdependent loads
308 only; it is not required to have any effect on stores, independent loads
309 or overlapping loads.
310
311 As mentioned in (1), the other CPUs in the system can be viewed as
312 committing sequences of stores to the memory system that the CPU being
313 considered can then perceive. A data dependency barrier issued by the CPU
314 under consideration guarantees that for any load preceding it, if that
315 load touches one of a sequence of stores from another CPU, then by the
316 time the barrier completes, the effects of all the stores prior to that
317 touched by the load will be perceptible to any loads issued after the data
318 dependency barrier.
319
320 See the "Examples of memory barrier sequences" subsection for diagrams
321 showing the ordering constraints.
322
323 [!] Note that the first load really has to have a _data_ dependency and
324 not a control dependency. If the address for the second load is dependent
325 on the first load, but the dependency is through a conditional rather than
326 actually loading the address itself, then it's a _control_ dependency and
327 a full read barrier or better is required. See the "Control dependencies"
328 subsection for more information.
329
330 [!] Note that data dependency barriers should normally be paired with
331 write barriers; see the "SMP barrier pairing" subsection.
332
333
334 (3) Read (or load) memory barriers.
335
336 A read barrier is a data dependency barrier plus a guarantee that all the
337 LOAD operations specified before the barrier will appear to happen before
338 all the LOAD operations specified after the barrier with respect to the
339 other components of the system.
340
341 A read barrier is a partial ordering on loads only; it is not required to
342 have any effect on stores.
343
344 Read memory barriers imply data dependency barriers, and so can substitute
345 for them.
346
347 [!] Note that read barriers should normally be paired with write barriers;
348 see the "SMP barrier pairing" subsection.
349
350
351 (4) General memory barriers.
352
David Howells670bd952006-06-10 09:54:12 -0700353 A general memory barrier gives a guarantee that all the LOAD and STORE
354 operations specified before the barrier will appear to happen before all
355 the LOAD and STORE operations specified after the barrier with respect to
356 the other components of the system.
357
358 A general memory barrier is a partial ordering over both loads and stores.
David Howells108b42b2006-03-31 16:00:29 +0100359
360 General memory barriers imply both read and write memory barriers, and so
361 can substitute for either.
362
363
364And a couple of implicit varieties:
365
366 (5) LOCK operations.
367
368 This acts as a one-way permeable barrier. It guarantees that all memory
369 operations after the LOCK operation will appear to happen after the LOCK
370 operation with respect to the other components of the system.
371
372 Memory operations that occur before a LOCK operation may appear to happen
373 after it completes.
374
375 A LOCK operation should almost always be paired with an UNLOCK operation.
376
377
378 (6) UNLOCK operations.
379
380 This also acts as a one-way permeable barrier. It guarantees that all
381 memory operations before the UNLOCK operation will appear to happen before
382 the UNLOCK operation with respect to the other components of the system.
383
384 Memory operations that occur after an UNLOCK operation may appear to
385 happen before it completes.
386
387 LOCK and UNLOCK operations are guaranteed to appear with respect to each
388 other strictly in the order specified.
389
390 The use of LOCK and UNLOCK operations generally precludes the need for
391 other sorts of memory barrier (but note the exceptions mentioned in the
392 subsection "MMIO write barrier").
393
394
395Memory barriers are only required where there's a possibility of interaction
396between two CPUs or between a CPU and a device. If it can be guaranteed that
397there won't be any such interaction in any particular piece of code, then
398memory barriers are unnecessary in that piece of code.
399
400
401Note that these are the _minimum_ guarantees. Different architectures may give
402more substantial guarantees, but they may _not_ be relied upon outside of arch
403specific code.
404
405
406WHAT MAY NOT BE ASSUMED ABOUT MEMORY BARRIERS?
407----------------------------------------------
408
409There are certain things that the Linux kernel memory barriers do not guarantee:
410
411 (*) There is no guarantee that any of the memory accesses specified before a
412 memory barrier will be _complete_ by the completion of a memory barrier
413 instruction; the barrier can be considered to draw a line in that CPU's
414 access queue that accesses of the appropriate type may not cross.
415
416 (*) There is no guarantee that issuing a memory barrier on one CPU will have
417 any direct effect on another CPU or any other hardware in the system. The
418 indirect effect will be the order in which the second CPU sees the effects
419 of the first CPU's accesses occur, but see the next point:
420
David Howells6bc39272006-06-25 05:49:22 -0700421 (*) There is no guarantee that a CPU will see the correct order of effects
David Howells108b42b2006-03-31 16:00:29 +0100422 from a second CPU's accesses, even _if_ the second CPU uses a memory
423 barrier, unless the first CPU _also_ uses a matching memory barrier (see
424 the subsection on "SMP Barrier Pairing").
425
426 (*) There is no guarantee that some intervening piece of off-the-CPU
427 hardware[*] will not reorder the memory accesses. CPU cache coherency
428 mechanisms should propagate the indirect effects of a memory barrier
429 between CPUs, but might not do so in order.
430
431 [*] For information on bus mastering DMA and coherency please read:
432
433 Documentation/pci.txt
434 Documentation/DMA-mapping.txt
435 Documentation/DMA-API.txt
436
437
438DATA DEPENDENCY BARRIERS
439------------------------
440
441The usage requirements of data dependency barriers are a little subtle, and
442it's not always obvious that they're needed. To illustrate, consider the
443following sequence of events:
444
445 CPU 1 CPU 2
446 =============== ===============
447 { A == 1, B == 2, C = 3, P == &A, Q == &C }
448 B = 4;
449 <write barrier>
450 P = &B
451 Q = P;
452 D = *Q;
453
454There's a clear data dependency here, and it would seem that by the end of the
455sequence, Q must be either &A or &B, and that:
456
457 (Q == &A) implies (D == 1)
458 (Q == &B) implies (D == 4)
459
Jarek Poplawski81fc6322007-05-23 13:58:20 -0700460But! CPU 2's perception of P may be updated _before_ its perception of B, thus
David Howells108b42b2006-03-31 16:00:29 +0100461leading to the following situation:
462
463 (Q == &B) and (D == 2) ????
464
465Whilst this may seem like a failure of coherency or causality maintenance, it
466isn't, and this behaviour can be observed on certain real CPUs (such as the DEC
467Alpha).
468
David Howells2b948952006-06-25 05:48:49 -0700469To deal with this, a data dependency barrier or better must be inserted
470between the address load and the data load:
David Howells108b42b2006-03-31 16:00:29 +0100471
472 CPU 1 CPU 2
473 =============== ===============
474 { A == 1, B == 2, C = 3, P == &A, Q == &C }
475 B = 4;
476 <write barrier>
477 P = &B
478 Q = P;
479 <data dependency barrier>
480 D = *Q;
481
482This enforces the occurrence of one of the two implications, and prevents the
483third possibility from arising.
484
485[!] Note that this extremely counterintuitive situation arises most easily on
486machines with split caches, so that, for example, one cache bank processes
487even-numbered cache lines and the other bank processes odd-numbered cache
488lines. The pointer P might be stored in an odd-numbered cache line, and the
489variable B might be stored in an even-numbered cache line. Then, if the
490even-numbered bank of the reading CPU's cache is extremely busy while the
491odd-numbered bank is idle, one can see the new value of the pointer P (&B),
David Howells6bc39272006-06-25 05:49:22 -0700492but the old value of the variable B (2).
David Howells108b42b2006-03-31 16:00:29 +0100493
494
495Another example of where data dependency barriers might by required is where a
496number is read from memory and then used to calculate the index for an array
497access:
498
499 CPU 1 CPU 2
500 =============== ===============
501 { M[0] == 1, M[1] == 2, M[3] = 3, P == 0, Q == 3 }
502 M[1] = 4;
503 <write barrier>
504 P = 1
505 Q = P;
506 <data dependency barrier>
507 D = M[Q];
508
509
510The data dependency barrier is very important to the RCU system, for example.
511See rcu_dereference() in include/linux/rcupdate.h. This permits the current
512target of an RCU'd pointer to be replaced with a new modified target, without
513the replacement target appearing to be incompletely initialised.
514
515See also the subsection on "Cache Coherency" for a more thorough example.
516
517
518CONTROL DEPENDENCIES
519--------------------
520
521A control dependency requires a full read memory barrier, not simply a data
522dependency barrier to make it work correctly. Consider the following bit of
523code:
524
525 q = &a;
526 if (p)
527 q = &b;
528 <data dependency barrier>
529 x = *q;
530
531This will not have the desired effect because there is no actual data
532dependency, but rather a control dependency that the CPU may short-circuit by
533attempting to predict the outcome in advance. In such a case what's actually
534required is:
535
536 q = &a;
537 if (p)
538 q = &b;
539 <read barrier>
540 x = *q;
541
542
543SMP BARRIER PAIRING
544-------------------
545
546When dealing with CPU-CPU interactions, certain types of memory barrier should
547always be paired. A lack of appropriate pairing is almost certainly an error.
548
549A write barrier should always be paired with a data dependency barrier or read
550barrier, though a general barrier would also be viable. Similarly a read
551barrier or a data dependency barrier should always be paired with at least an
552write barrier, though, again, a general barrier is viable:
553
554 CPU 1 CPU 2
555 =============== ===============
556 a = 1;
557 <write barrier>
David Howells670bd952006-06-10 09:54:12 -0700558 b = 2; x = b;
David Howells108b42b2006-03-31 16:00:29 +0100559 <read barrier>
David Howells670bd952006-06-10 09:54:12 -0700560 y = a;
David Howells108b42b2006-03-31 16:00:29 +0100561
562Or:
563
564 CPU 1 CPU 2
565 =============== ===============================
566 a = 1;
567 <write barrier>
568 b = &a; x = b;
569 <data dependency barrier>
570 y = *x;
571
572Basically, the read barrier always has to be there, even though it can be of
573the "weaker" type.
574
David Howells670bd952006-06-10 09:54:12 -0700575[!] Note that the stores before the write barrier would normally be expected to
Jarek Poplawski81fc6322007-05-23 13:58:20 -0700576match the loads after the read barrier or the data dependency barrier, and vice
David Howells670bd952006-06-10 09:54:12 -0700577versa:
578
579 CPU 1 CPU 2
580 =============== ===============
581 a = 1; }---- --->{ v = c
582 b = 2; } \ / { w = d
583 <write barrier> \ <read barrier>
584 c = 3; } / \ { x = a;
585 d = 4; }---- --->{ y = b;
586
David Howells108b42b2006-03-31 16:00:29 +0100587
588EXAMPLES OF MEMORY BARRIER SEQUENCES
589------------------------------------
590
Jarek Poplawski81fc6322007-05-23 13:58:20 -0700591Firstly, write barriers act as partial orderings on store operations.
David Howells108b42b2006-03-31 16:00:29 +0100592Consider the following sequence of events:
593
594 CPU 1
595 =======================
596 STORE A = 1
597 STORE B = 2
598 STORE C = 3
599 <write barrier>
600 STORE D = 4
601 STORE E = 5
602
603This sequence of events is committed to the memory coherence system in an order
604that the rest of the system might perceive as the unordered set of { STORE A,
Adrian Bunk80f72282006-06-30 18:27:16 +0200605STORE B, STORE C } all occurring before the unordered set of { STORE D, STORE E
David Howells108b42b2006-03-31 16:00:29 +0100606}:
607
608 +-------+ : :
609 | | +------+
610 | |------>| C=3 | } /\
Jarek Poplawski81fc6322007-05-23 13:58:20 -0700611 | | : +------+ }----- \ -----> Events perceptible to
612 | | : | A=1 | } \/ the rest of the system
David Howells108b42b2006-03-31 16:00:29 +0100613 | | : +------+ }
614 | CPU 1 | : | B=2 | }
615 | | +------+ }
616 | | wwwwwwwwwwwwwwww } <--- At this point the write barrier
617 | | +------+ } requires all stores prior to the
618 | | : | E=5 | } barrier to be committed before
Jarek Poplawski81fc6322007-05-23 13:58:20 -0700619 | | : +------+ } further stores may take place
David Howells108b42b2006-03-31 16:00:29 +0100620 | |------>| D=4 | }
621 | | +------+
622 +-------+ : :
623 |
David Howells670bd952006-06-10 09:54:12 -0700624 | Sequence in which stores are committed to the
625 | memory system by CPU 1
David Howells108b42b2006-03-31 16:00:29 +0100626 V
627
628
Jarek Poplawski81fc6322007-05-23 13:58:20 -0700629Secondly, data dependency barriers act as partial orderings on data-dependent
David Howells108b42b2006-03-31 16:00:29 +0100630loads. Consider the following sequence of events:
631
632 CPU 1 CPU 2
633 ======================= =======================
David Howellsc14038c2006-04-10 22:54:24 -0700634 { B = 7; X = 9; Y = 8; C = &Y }
David Howells108b42b2006-03-31 16:00:29 +0100635 STORE A = 1
636 STORE B = 2
637 <write barrier>
638 STORE C = &B LOAD X
639 STORE D = 4 LOAD C (gets &B)
640 LOAD *C (reads B)
641
642Without intervention, CPU 2 may perceive the events on CPU 1 in some
643effectively random order, despite the write barrier issued by CPU 1:
644
645 +-------+ : : : :
646 | | +------+ +-------+ | Sequence of update
647 | |------>| B=2 |----- --->| Y->8 | | of perception on
648 | | : +------+ \ +-------+ | CPU 2
649 | CPU 1 | : | A=1 | \ --->| C->&Y | V
650 | | +------+ | +-------+
651 | | wwwwwwwwwwwwwwww | : :
652 | | +------+ | : :
653 | | : | C=&B |--- | : : +-------+
654 | | : +------+ \ | +-------+ | |
655 | |------>| D=4 | ----------->| C->&B |------>| |
656 | | +------+ | +-------+ | |
657 +-------+ : : | : : | |
658 | : : | |
659 | : : | CPU 2 |
660 | +-------+ | |
661 Apparently incorrect ---> | | B->7 |------>| |
662 perception of B (!) | +-------+ | |
663 | : : | |
664 | +-------+ | |
665 The load of X holds ---> \ | X->9 |------>| |
666 up the maintenance \ +-------+ | |
667 of coherence of B ----->| B->2 | +-------+
668 +-------+
669 : :
670
671
672In the above example, CPU 2 perceives that B is 7, despite the load of *C
Paolo Ornati670e9f32006-10-03 22:57:56 +0200673(which would be B) coming after the LOAD of C.
David Howells108b42b2006-03-31 16:00:29 +0100674
675If, however, a data dependency barrier were to be placed between the load of C
David Howellsc14038c2006-04-10 22:54:24 -0700676and the load of *C (ie: B) on CPU 2:
677
678 CPU 1 CPU 2
679 ======================= =======================
680 { B = 7; X = 9; Y = 8; C = &Y }
681 STORE A = 1
682 STORE B = 2
683 <write barrier>
684 STORE C = &B LOAD X
685 STORE D = 4 LOAD C (gets &B)
686 <data dependency barrier>
687 LOAD *C (reads B)
688
689then the following will occur:
David Howells108b42b2006-03-31 16:00:29 +0100690
691 +-------+ : : : :
692 | | +------+ +-------+
693 | |------>| B=2 |----- --->| Y->8 |
694 | | : +------+ \ +-------+
695 | CPU 1 | : | A=1 | \ --->| C->&Y |
696 | | +------+ | +-------+
697 | | wwwwwwwwwwwwwwww | : :
698 | | +------+ | : :
699 | | : | C=&B |--- | : : +-------+
700 | | : +------+ \ | +-------+ | |
701 | |------>| D=4 | ----------->| C->&B |------>| |
702 | | +------+ | +-------+ | |
703 +-------+ : : | : : | |
704 | : : | |
705 | : : | CPU 2 |
706 | +-------+ | |
David Howells670bd952006-06-10 09:54:12 -0700707 | | X->9 |------>| |
708 | +-------+ | |
709 Makes sure all effects ---> \ ddddddddddddddddd | |
710 prior to the store of C \ +-------+ | |
711 are perceptible to ----->| B->2 |------>| |
712 subsequent loads +-------+ | |
David Howells108b42b2006-03-31 16:00:29 +0100713 : : +-------+
714
715
716And thirdly, a read barrier acts as a partial order on loads. Consider the
717following sequence of events:
718
719 CPU 1 CPU 2
720 ======================= =======================
David Howells670bd952006-06-10 09:54:12 -0700721 { A = 0, B = 9 }
David Howells108b42b2006-03-31 16:00:29 +0100722 STORE A=1
David Howells108b42b2006-03-31 16:00:29 +0100723 <write barrier>
David Howells670bd952006-06-10 09:54:12 -0700724 STORE B=2
David Howells108b42b2006-03-31 16:00:29 +0100725 LOAD B
David Howells670bd952006-06-10 09:54:12 -0700726 LOAD A
David Howells108b42b2006-03-31 16:00:29 +0100727
728Without intervention, CPU 2 may then choose to perceive the events on CPU 1 in
729some effectively random order, despite the write barrier issued by CPU 1:
730
David Howells670bd952006-06-10 09:54:12 -0700731 +-------+ : : : :
732 | | +------+ +-------+
733 | |------>| A=1 |------ --->| A->0 |
734 | | +------+ \ +-------+
735 | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 |
736 | | +------+ | +-------+
737 | |------>| B=2 |--- | : :
738 | | +------+ \ | : : +-------+
739 +-------+ : : \ | +-------+ | |
740 ---------->| B->2 |------>| |
741 | +-------+ | CPU 2 |
742 | | A->0 |------>| |
743 | +-------+ | |
744 | : : +-------+
745 \ : :
746 \ +-------+
747 ---->| A->1 |
748 +-------+
749 : :
David Howells108b42b2006-03-31 16:00:29 +0100750
751
David Howells6bc39272006-06-25 05:49:22 -0700752If, however, a read barrier were to be placed between the load of B and the
David Howells670bd952006-06-10 09:54:12 -0700753load of A on CPU 2:
David Howells108b42b2006-03-31 16:00:29 +0100754
David Howells670bd952006-06-10 09:54:12 -0700755 CPU 1 CPU 2
756 ======================= =======================
757 { A = 0, B = 9 }
758 STORE A=1
759 <write barrier>
760 STORE B=2
761 LOAD B
762 <read barrier>
763 LOAD A
764
765then the partial ordering imposed by CPU 1 will be perceived correctly by CPU
7662:
767
768 +-------+ : : : :
769 | | +------+ +-------+
770 | |------>| A=1 |------ --->| A->0 |
771 | | +------+ \ +-------+
772 | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 |
773 | | +------+ | +-------+
774 | |------>| B=2 |--- | : :
775 | | +------+ \ | : : +-------+
776 +-------+ : : \ | +-------+ | |
777 ---------->| B->2 |------>| |
778 | +-------+ | CPU 2 |
779 | : : | |
780 | : : | |
781 At this point the read ----> \ rrrrrrrrrrrrrrrrr | |
782 barrier causes all effects \ +-------+ | |
783 prior to the storage of B ---->| A->1 |------>| |
784 to be perceptible to CPU 2 +-------+ | |
785 : : +-------+
786
787
788To illustrate this more completely, consider what could happen if the code
789contained a load of A either side of the read barrier:
790
791 CPU 1 CPU 2
792 ======================= =======================
793 { A = 0, B = 9 }
794 STORE A=1
795 <write barrier>
796 STORE B=2
797 LOAD B
798 LOAD A [first load of A]
799 <read barrier>
800 LOAD A [second load of A]
801
802Even though the two loads of A both occur after the load of B, they may both
803come up with different values:
804
805 +-------+ : : : :
806 | | +------+ +-------+
807 | |------>| A=1 |------ --->| A->0 |
808 | | +------+ \ +-------+
809 | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 |
810 | | +------+ | +-------+
811 | |------>| B=2 |--- | : :
812 | | +------+ \ | : : +-------+
813 +-------+ : : \ | +-------+ | |
814 ---------->| B->2 |------>| |
815 | +-------+ | CPU 2 |
816 | : : | |
817 | : : | |
818 | +-------+ | |
819 | | A->0 |------>| 1st |
820 | +-------+ | |
821 At this point the read ----> \ rrrrrrrrrrrrrrrrr | |
822 barrier causes all effects \ +-------+ | |
823 prior to the storage of B ---->| A->1 |------>| 2nd |
824 to be perceptible to CPU 2 +-------+ | |
825 : : +-------+
826
827
828But it may be that the update to A from CPU 1 becomes perceptible to CPU 2
829before the read barrier completes anyway:
830
831 +-------+ : : : :
832 | | +------+ +-------+
833 | |------>| A=1 |------ --->| A->0 |
834 | | +------+ \ +-------+
835 | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 |
836 | | +------+ | +-------+
837 | |------>| B=2 |--- | : :
838 | | +------+ \ | : : +-------+
839 +-------+ : : \ | +-------+ | |
840 ---------->| B->2 |------>| |
841 | +-------+ | CPU 2 |
842 | : : | |
843 \ : : | |
844 \ +-------+ | |
845 ---->| A->1 |------>| 1st |
846 +-------+ | |
847 rrrrrrrrrrrrrrrrr | |
848 +-------+ | |
849 | A->1 |------>| 2nd |
850 +-------+ | |
851 : : +-------+
852
853
854The guarantee is that the second load will always come up with A == 1 if the
855load of B came up with B == 2. No such guarantee exists for the first load of
856A; that may come up with either A == 0 or A == 1.
857
858
859READ MEMORY BARRIERS VS LOAD SPECULATION
860----------------------------------------
861
862Many CPUs speculate with loads: that is they see that they will need to load an
863item from memory, and they find a time where they're not using the bus for any
864other loads, and so do the load in advance - even though they haven't actually
865got to that point in the instruction execution flow yet. This permits the
866actual load instruction to potentially complete immediately because the CPU
867already has the value to hand.
868
869It may turn out that the CPU didn't actually need the value - perhaps because a
870branch circumvented the load - in which case it can discard the value or just
871cache it for later use.
872
873Consider:
874
875 CPU 1 CPU 2
876 ======================= =======================
877 LOAD B
878 DIVIDE } Divide instructions generally
879 DIVIDE } take a long time to perform
880 LOAD A
881
882Which might appear as this:
883
884 : : +-------+
885 +-------+ | |
886 --->| B->2 |------>| |
887 +-------+ | CPU 2 |
888 : :DIVIDE | |
889 +-------+ | |
890 The CPU being busy doing a ---> --->| A->0 |~~~~ | |
891 division speculates on the +-------+ ~ | |
892 LOAD of A : : ~ | |
893 : :DIVIDE | |
894 : : ~ | |
895 Once the divisions are complete --> : : ~-->| |
896 the CPU can then perform the : : | |
897 LOAD with immediate effect : : +-------+
898
899
900Placing a read barrier or a data dependency barrier just before the second
901load:
902
903 CPU 1 CPU 2
904 ======================= =======================
905 LOAD B
906 DIVIDE
907 DIVIDE
908 <read barrier>
909 LOAD A
910
911will force any value speculatively obtained to be reconsidered to an extent
912dependent on the type of barrier used. If there was no change made to the
913speculated memory location, then the speculated value will just be used:
914
915 : : +-------+
916 +-------+ | |
917 --->| B->2 |------>| |
918 +-------+ | CPU 2 |
919 : :DIVIDE | |
920 +-------+ | |
921 The CPU being busy doing a ---> --->| A->0 |~~~~ | |
922 division speculates on the +-------+ ~ | |
923 LOAD of A : : ~ | |
924 : :DIVIDE | |
925 : : ~ | |
926 : : ~ | |
927 rrrrrrrrrrrrrrrr~ | |
928 : : ~ | |
929 : : ~-->| |
930 : : | |
931 : : +-------+
932
933
934but if there was an update or an invalidation from another CPU pending, then
935the speculation will be cancelled and the value reloaded:
936
937 : : +-------+
938 +-------+ | |
939 --->| B->2 |------>| |
940 +-------+ | CPU 2 |
941 : :DIVIDE | |
942 +-------+ | |
943 The CPU being busy doing a ---> --->| A->0 |~~~~ | |
944 division speculates on the +-------+ ~ | |
945 LOAD of A : : ~ | |
946 : :DIVIDE | |
947 : : ~ | |
948 : : ~ | |
949 rrrrrrrrrrrrrrrrr | |
950 +-------+ | |
951 The speculation is discarded ---> --->| A->1 |------>| |
952 and an updated value is +-------+ | |
953 retrieved : : +-------+
David Howells108b42b2006-03-31 16:00:29 +0100954
955
956========================
957EXPLICIT KERNEL BARRIERS
958========================
959
960The Linux kernel has a variety of different barriers that act at different
961levels:
962
963 (*) Compiler barrier.
964
965 (*) CPU memory barriers.
966
967 (*) MMIO write barrier.
968
969
970COMPILER BARRIER
971----------------
972
973The Linux kernel has an explicit compiler barrier function that prevents the
974compiler from moving the memory accesses either side of it to the other side:
975
976 barrier();
977
Jarek Poplawski81fc6322007-05-23 13:58:20 -0700978This is a general barrier - lesser varieties of compiler barrier do not exist.
David Howells108b42b2006-03-31 16:00:29 +0100979
980The compiler barrier has no direct effect on the CPU, which may then reorder
981things however it wishes.
982
983
984CPU MEMORY BARRIERS
985-------------------
986
987The Linux kernel has eight basic CPU memory barriers:
988
989 TYPE MANDATORY SMP CONDITIONAL
990 =============== ======================= ===========================
991 GENERAL mb() smp_mb()
992 WRITE wmb() smp_wmb()
993 READ rmb() smp_rmb()
994 DATA DEPENDENCY read_barrier_depends() smp_read_barrier_depends()
995
996
997All CPU memory barriers unconditionally imply compiler barriers.
998
999SMP memory barriers are reduced to compiler barriers on uniprocessor compiled
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001000systems because it is assumed that a CPU will appear to be self-consistent,
David Howells108b42b2006-03-31 16:00:29 +01001001and will order overlapping accesses correctly with respect to itself.
1002
1003[!] Note that SMP memory barriers _must_ be used to control the ordering of
1004references to shared memory on SMP systems, though the use of locking instead
1005is sufficient.
1006
1007Mandatory barriers should not be used to control SMP effects, since mandatory
1008barriers unnecessarily impose overhead on UP systems. They may, however, be
1009used to control MMIO effects on accesses through relaxed memory I/O windows.
1010These are required even on non-SMP systems as they affect the order in which
1011memory operations appear to a device by prohibiting both the compiler and the
1012CPU from reordering them.
1013
1014
1015There are some more advanced barrier functions:
1016
1017 (*) set_mb(var, value)
David Howells108b42b2006-03-31 16:00:29 +01001018
Oleg Nesterov75b2bd52006-11-08 17:44:38 -08001019 This assigns the value to the variable and then inserts a full memory
Steven Rostedtf92213b2006-07-14 16:05:01 -04001020 barrier after it, depending on the function. It isn't guaranteed to
David Howells108b42b2006-03-31 16:00:29 +01001021 insert anything more than a compiler barrier in a UP compilation.
1022
1023
1024 (*) smp_mb__before_atomic_dec();
1025 (*) smp_mb__after_atomic_dec();
1026 (*) smp_mb__before_atomic_inc();
1027 (*) smp_mb__after_atomic_inc();
1028
1029 These are for use with atomic add, subtract, increment and decrement
David Howellsdbc87002006-04-10 22:54:23 -07001030 functions that don't return a value, especially when used for reference
1031 counting. These functions do not imply memory barriers.
David Howells108b42b2006-03-31 16:00:29 +01001032
1033 As an example, consider a piece of code that marks an object as being dead
1034 and then decrements the object's reference count:
1035
1036 obj->dead = 1;
1037 smp_mb__before_atomic_dec();
1038 atomic_dec(&obj->ref_count);
1039
1040 This makes sure that the death mark on the object is perceived to be set
1041 *before* the reference counter is decremented.
1042
1043 See Documentation/atomic_ops.txt for more information. See the "Atomic
1044 operations" subsection for information on where to use these.
1045
1046
1047 (*) smp_mb__before_clear_bit(void);
1048 (*) smp_mb__after_clear_bit(void);
1049
1050 These are for use similar to the atomic inc/dec barriers. These are
1051 typically used for bitwise unlocking operations, so care must be taken as
1052 there are no implicit memory barriers here either.
1053
1054 Consider implementing an unlock operation of some nature by clearing a
1055 locking bit. The clear_bit() would then need to be barriered like this:
1056
1057 smp_mb__before_clear_bit();
1058 clear_bit( ... );
1059
1060 This prevents memory operations before the clear leaking to after it. See
1061 the subsection on "Locking Functions" with reference to UNLOCK operation
1062 implications.
1063
1064 See Documentation/atomic_ops.txt for more information. See the "Atomic
1065 operations" subsection for information on where to use these.
1066
1067
1068MMIO WRITE BARRIER
1069------------------
1070
1071The Linux kernel also has a special barrier for use with memory-mapped I/O
1072writes:
1073
1074 mmiowb();
1075
1076This is a variation on the mandatory write barrier that causes writes to weakly
1077ordered I/O regions to be partially ordered. Its effects may go beyond the
1078CPU->Hardware interface and actually affect the hardware at some level.
1079
1080See the subsection "Locks vs I/O accesses" for more information.
1081
1082
1083===============================
1084IMPLICIT KERNEL MEMORY BARRIERS
1085===============================
1086
1087Some of the other functions in the linux kernel imply memory barriers, amongst
David Howells670bd952006-06-10 09:54:12 -07001088which are locking and scheduling functions.
David Howells108b42b2006-03-31 16:00:29 +01001089
1090This specification is a _minimum_ guarantee; any particular architecture may
1091provide more substantial guarantees, but these may not be relied upon outside
1092of arch specific code.
1093
1094
1095LOCKING FUNCTIONS
1096-----------------
1097
1098The Linux kernel has a number of locking constructs:
1099
1100 (*) spin locks
1101 (*) R/W spin locks
1102 (*) mutexes
1103 (*) semaphores
1104 (*) R/W semaphores
1105 (*) RCU
1106
1107In all cases there are variants on "LOCK" operations and "UNLOCK" operations
1108for each construct. These operations all imply certain barriers:
1109
1110 (1) LOCK operation implication:
1111
1112 Memory operations issued after the LOCK will be completed after the LOCK
1113 operation has completed.
1114
1115 Memory operations issued before the LOCK may be completed after the LOCK
1116 operation has completed.
1117
1118 (2) UNLOCK operation implication:
1119
1120 Memory operations issued before the UNLOCK will be completed before the
1121 UNLOCK operation has completed.
1122
1123 Memory operations issued after the UNLOCK may be completed before the
1124 UNLOCK operation has completed.
1125
1126 (3) LOCK vs LOCK implication:
1127
1128 All LOCK operations issued before another LOCK operation will be completed
1129 before that LOCK operation.
1130
1131 (4) LOCK vs UNLOCK implication:
1132
1133 All LOCK operations issued before an UNLOCK operation will be completed
1134 before the UNLOCK operation.
1135
1136 All UNLOCK operations issued before a LOCK operation will be completed
1137 before the LOCK operation.
1138
1139 (5) Failed conditional LOCK implication:
1140
1141 Certain variants of the LOCK operation may fail, either due to being
1142 unable to get the lock immediately, or due to receiving an unblocked
1143 signal whilst asleep waiting for the lock to become available. Failed
1144 locks do not imply any sort of barrier.
1145
1146Therefore, from (1), (2) and (4) an UNLOCK followed by an unconditional LOCK is
1147equivalent to a full barrier, but a LOCK followed by an UNLOCK is not.
1148
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001149[!] Note: one of the consequences of LOCKs and UNLOCKs being only one-way
1150 barriers is that the effects of instructions outside of a critical section
1151 may seep into the inside of the critical section.
David Howells108b42b2006-03-31 16:00:29 +01001152
David Howells670bd952006-06-10 09:54:12 -07001153A LOCK followed by an UNLOCK may not be assumed to be full memory barrier
1154because it is possible for an access preceding the LOCK to happen after the
1155LOCK, and an access following the UNLOCK to happen before the UNLOCK, and the
1156two accesses can themselves then cross:
1157
1158 *A = a;
1159 LOCK
1160 UNLOCK
1161 *B = b;
1162
1163may occur as:
1164
1165 LOCK, STORE *B, STORE *A, UNLOCK
1166
David Howells108b42b2006-03-31 16:00:29 +01001167Locks and semaphores may not provide any guarantee of ordering on UP compiled
1168systems, and so cannot be counted on in such a situation to actually achieve
1169anything at all - especially with respect to I/O accesses - unless combined
1170with interrupt disabling operations.
1171
1172See also the section on "Inter-CPU locking barrier effects".
1173
1174
1175As an example, consider the following:
1176
1177 *A = a;
1178 *B = b;
1179 LOCK
1180 *C = c;
1181 *D = d;
1182 UNLOCK
1183 *E = e;
1184 *F = f;
1185
1186The following sequence of events is acceptable:
1187
1188 LOCK, {*F,*A}, *E, {*C,*D}, *B, UNLOCK
1189
1190 [+] Note that {*F,*A} indicates a combined access.
1191
1192But none of the following are:
1193
1194 {*F,*A}, *B, LOCK, *C, *D, UNLOCK, *E
1195 *A, *B, *C, LOCK, *D, UNLOCK, *E, *F
1196 *A, *B, LOCK, *C, UNLOCK, *D, *E, *F
1197 *B, LOCK, *C, *D, UNLOCK, {*F,*A}, *E
1198
1199
1200
1201INTERRUPT DISABLING FUNCTIONS
1202-----------------------------
1203
1204Functions that disable interrupts (LOCK equivalent) and enable interrupts
1205(UNLOCK equivalent) will act as compiler barriers only. So if memory or I/O
1206barriers are required in such a situation, they must be provided from some
1207other means.
1208
1209
1210MISCELLANEOUS FUNCTIONS
1211-----------------------
1212
1213Other functions that imply barriers:
1214
1215 (*) schedule() and similar imply full memory barriers.
1216
David Howells108b42b2006-03-31 16:00:29 +01001217
1218=================================
1219INTER-CPU LOCKING BARRIER EFFECTS
1220=================================
1221
1222On SMP systems locking primitives give a more substantial form of barrier: one
1223that does affect memory access ordering on other CPUs, within the context of
1224conflict on any particular lock.
1225
1226
1227LOCKS VS MEMORY ACCESSES
1228------------------------
1229
Aneesh Kumar79afecf2006-05-15 09:44:36 -07001230Consider the following: the system has a pair of spinlocks (M) and (Q), and
David Howells108b42b2006-03-31 16:00:29 +01001231three CPUs; then should the following sequence of events occur:
1232
1233 CPU 1 CPU 2
1234 =============================== ===============================
1235 *A = a; *E = e;
1236 LOCK M LOCK Q
1237 *B = b; *F = f;
1238 *C = c; *G = g;
1239 UNLOCK M UNLOCK Q
1240 *D = d; *H = h;
1241
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001242Then there is no guarantee as to what order CPU 3 will see the accesses to *A
David Howells108b42b2006-03-31 16:00:29 +01001243through *H occur in, other than the constraints imposed by the separate locks
1244on the separate CPUs. It might, for example, see:
1245
1246 *E, LOCK M, LOCK Q, *G, *C, *F, *A, *B, UNLOCK Q, *D, *H, UNLOCK M
1247
1248But it won't see any of:
1249
1250 *B, *C or *D preceding LOCK M
1251 *A, *B or *C following UNLOCK M
1252 *F, *G or *H preceding LOCK Q
1253 *E, *F or *G following UNLOCK Q
1254
1255
1256However, if the following occurs:
1257
1258 CPU 1 CPU 2
1259 =============================== ===============================
1260 *A = a;
1261 LOCK M [1]
1262 *B = b;
1263 *C = c;
1264 UNLOCK M [1]
1265 *D = d; *E = e;
1266 LOCK M [2]
1267 *F = f;
1268 *G = g;
1269 UNLOCK M [2]
1270 *H = h;
1271
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001272CPU 3 might see:
David Howells108b42b2006-03-31 16:00:29 +01001273
1274 *E, LOCK M [1], *C, *B, *A, UNLOCK M [1],
1275 LOCK M [2], *H, *F, *G, UNLOCK M [2], *D
1276
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001277But assuming CPU 1 gets the lock first, CPU 3 won't see any of:
David Howells108b42b2006-03-31 16:00:29 +01001278
1279 *B, *C, *D, *F, *G or *H preceding LOCK M [1]
1280 *A, *B or *C following UNLOCK M [1]
1281 *F, *G or *H preceding LOCK M [2]
1282 *A, *B, *C, *E, *F or *G following UNLOCK M [2]
1283
1284
1285LOCKS VS I/O ACCESSES
1286---------------------
1287
1288Under certain circumstances (especially involving NUMA), I/O accesses within
1289two spinlocked sections on two different CPUs may be seen as interleaved by the
1290PCI bridge, because the PCI bridge does not necessarily participate in the
1291cache-coherence protocol, and is therefore incapable of issuing the required
1292read memory barriers.
1293
1294For example:
1295
1296 CPU 1 CPU 2
1297 =============================== ===============================
1298 spin_lock(Q)
1299 writel(0, ADDR)
1300 writel(1, DATA);
1301 spin_unlock(Q);
1302 spin_lock(Q);
1303 writel(4, ADDR);
1304 writel(5, DATA);
1305 spin_unlock(Q);
1306
1307may be seen by the PCI bridge as follows:
1308
1309 STORE *ADDR = 0, STORE *ADDR = 4, STORE *DATA = 1, STORE *DATA = 5
1310
1311which would probably cause the hardware to malfunction.
1312
1313
1314What is necessary here is to intervene with an mmiowb() before dropping the
1315spinlock, for example:
1316
1317 CPU 1 CPU 2
1318 =============================== ===============================
1319 spin_lock(Q)
1320 writel(0, ADDR)
1321 writel(1, DATA);
1322 mmiowb();
1323 spin_unlock(Q);
1324 spin_lock(Q);
1325 writel(4, ADDR);
1326 writel(5, DATA);
1327 mmiowb();
1328 spin_unlock(Q);
1329
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001330this will ensure that the two stores issued on CPU 1 appear at the PCI bridge
1331before either of the stores issued on CPU 2.
David Howells108b42b2006-03-31 16:00:29 +01001332
1333
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001334Furthermore, following a store by a load from the same device obviates the need
1335for the mmiowb(), because the load forces the store to complete before the load
David Howells108b42b2006-03-31 16:00:29 +01001336is performed:
1337
1338 CPU 1 CPU 2
1339 =============================== ===============================
1340 spin_lock(Q)
1341 writel(0, ADDR)
1342 a = readl(DATA);
1343 spin_unlock(Q);
1344 spin_lock(Q);
1345 writel(4, ADDR);
1346 b = readl(DATA);
1347 spin_unlock(Q);
1348
1349
1350See Documentation/DocBook/deviceiobook.tmpl for more information.
1351
1352
1353=================================
1354WHERE ARE MEMORY BARRIERS NEEDED?
1355=================================
1356
1357Under normal operation, memory operation reordering is generally not going to
1358be a problem as a single-threaded linear piece of code will still appear to
1359work correctly, even if it's in an SMP kernel. There are, however, three
1360circumstances in which reordering definitely _could_ be a problem:
1361
1362 (*) Interprocessor interaction.
1363
1364 (*) Atomic operations.
1365
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001366 (*) Accessing devices.
David Howells108b42b2006-03-31 16:00:29 +01001367
1368 (*) Interrupts.
1369
1370
1371INTERPROCESSOR INTERACTION
1372--------------------------
1373
1374When there's a system with more than one processor, more than one CPU in the
1375system may be working on the same data set at the same time. This can cause
1376synchronisation problems, and the usual way of dealing with them is to use
1377locks. Locks, however, are quite expensive, and so it may be preferable to
1378operate without the use of a lock if at all possible. In such a case
1379operations that affect both CPUs may have to be carefully ordered to prevent
1380a malfunction.
1381
1382Consider, for example, the R/W semaphore slow path. Here a waiting process is
1383queued on the semaphore, by virtue of it having a piece of its stack linked to
1384the semaphore's list of waiting processes:
1385
1386 struct rw_semaphore {
1387 ...
1388 spinlock_t lock;
1389 struct list_head waiters;
1390 };
1391
1392 struct rwsem_waiter {
1393 struct list_head list;
1394 struct task_struct *task;
1395 };
1396
1397To wake up a particular waiter, the up_read() or up_write() functions have to:
1398
1399 (1) read the next pointer from this waiter's record to know as to where the
1400 next waiter record is;
1401
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001402 (2) read the pointer to the waiter's task structure;
David Howells108b42b2006-03-31 16:00:29 +01001403
1404 (3) clear the task pointer to tell the waiter it has been given the semaphore;
1405
1406 (4) call wake_up_process() on the task; and
1407
1408 (5) release the reference held on the waiter's task struct.
1409
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001410In other words, it has to perform this sequence of events:
David Howells108b42b2006-03-31 16:00:29 +01001411
1412 LOAD waiter->list.next;
1413 LOAD waiter->task;
1414 STORE waiter->task;
1415 CALL wakeup
1416 RELEASE task
1417
1418and if any of these steps occur out of order, then the whole thing may
1419malfunction.
1420
1421Once it has queued itself and dropped the semaphore lock, the waiter does not
1422get the lock again; it instead just waits for its task pointer to be cleared
1423before proceeding. Since the record is on the waiter's stack, this means that
1424if the task pointer is cleared _before_ the next pointer in the list is read,
1425another CPU might start processing the waiter and might clobber the waiter's
1426stack before the up*() function has a chance to read the next pointer.
1427
1428Consider then what might happen to the above sequence of events:
1429
1430 CPU 1 CPU 2
1431 =============================== ===============================
1432 down_xxx()
1433 Queue waiter
1434 Sleep
1435 up_yyy()
1436 LOAD waiter->task;
1437 STORE waiter->task;
1438 Woken up by other event
1439 <preempt>
1440 Resume processing
1441 down_xxx() returns
1442 call foo()
1443 foo() clobbers *waiter
1444 </preempt>
1445 LOAD waiter->list.next;
1446 --- OOPS ---
1447
1448This could be dealt with using the semaphore lock, but then the down_xxx()
1449function has to needlessly get the spinlock again after being woken up.
1450
1451The way to deal with this is to insert a general SMP memory barrier:
1452
1453 LOAD waiter->list.next;
1454 LOAD waiter->task;
1455 smp_mb();
1456 STORE waiter->task;
1457 CALL wakeup
1458 RELEASE task
1459
1460In this case, the barrier makes a guarantee that all memory accesses before the
1461barrier will appear to happen before all the memory accesses after the barrier
1462with respect to the other CPUs on the system. It does _not_ guarantee that all
1463the memory accesses before the barrier will be complete by the time the barrier
1464instruction itself is complete.
1465
1466On a UP system - where this wouldn't be a problem - the smp_mb() is just a
1467compiler barrier, thus making sure the compiler emits the instructions in the
David Howells6bc39272006-06-25 05:49:22 -07001468right order without actually intervening in the CPU. Since there's only one
1469CPU, that CPU's dependency ordering logic will take care of everything else.
David Howells108b42b2006-03-31 16:00:29 +01001470
1471
1472ATOMIC OPERATIONS
1473-----------------
1474
David Howellsdbc87002006-04-10 22:54:23 -07001475Whilst they are technically interprocessor interaction considerations, atomic
1476operations are noted specially as some of them imply full memory barriers and
1477some don't, but they're very heavily relied on as a group throughout the
1478kernel.
1479
1480Any atomic operation that modifies some state in memory and returns information
1481about the state (old or new) implies an SMP-conditional general memory barrier
1482(smp_mb()) on each side of the actual operation. These include:
David Howells108b42b2006-03-31 16:00:29 +01001483
1484 xchg();
1485 cmpxchg();
David Howells108b42b2006-03-31 16:00:29 +01001486 atomic_cmpxchg();
1487 atomic_inc_return();
1488 atomic_dec_return();
1489 atomic_add_return();
1490 atomic_sub_return();
1491 atomic_inc_and_test();
1492 atomic_dec_and_test();
1493 atomic_sub_and_test();
1494 atomic_add_negative();
1495 atomic_add_unless();
David Howellsdbc87002006-04-10 22:54:23 -07001496 test_and_set_bit();
1497 test_and_clear_bit();
1498 test_and_change_bit();
David Howells108b42b2006-03-31 16:00:29 +01001499
David Howellsdbc87002006-04-10 22:54:23 -07001500These are used for such things as implementing LOCK-class and UNLOCK-class
1501operations and adjusting reference counters towards object destruction, and as
1502such the implicit memory barrier effects are necessary.
David Howells108b42b2006-03-31 16:00:29 +01001503
David Howells108b42b2006-03-31 16:00:29 +01001504
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001505The following operations are potential problems as they do _not_ imply memory
David Howellsdbc87002006-04-10 22:54:23 -07001506barriers, but might be used for implementing such things as UNLOCK-class
1507operations:
1508
1509 atomic_set();
David Howells108b42b2006-03-31 16:00:29 +01001510 set_bit();
1511 clear_bit();
1512 change_bit();
David Howellsdbc87002006-04-10 22:54:23 -07001513
1514With these the appropriate explicit memory barrier should be used if necessary
1515(smp_mb__before_clear_bit() for instance).
David Howells108b42b2006-03-31 16:00:29 +01001516
1517
David Howellsdbc87002006-04-10 22:54:23 -07001518The following also do _not_ imply memory barriers, and so may require explicit
1519memory barriers under some circumstances (smp_mb__before_atomic_dec() for
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001520instance):
David Howells108b42b2006-03-31 16:00:29 +01001521
1522 atomic_add();
1523 atomic_sub();
1524 atomic_inc();
1525 atomic_dec();
1526
1527If they're used for statistics generation, then they probably don't need memory
1528barriers, unless there's a coupling between statistical data.
1529
1530If they're used for reference counting on an object to control its lifetime,
1531they probably don't need memory barriers because either the reference count
1532will be adjusted inside a locked section, or the caller will already hold
1533sufficient references to make the lock, and thus a memory barrier unnecessary.
1534
1535If they're used for constructing a lock of some description, then they probably
1536do need memory barriers as a lock primitive generally has to do things in a
1537specific order.
1538
1539
1540Basically, each usage case has to be carefully considered as to whether memory
David Howellsdbc87002006-04-10 22:54:23 -07001541barriers are needed or not.
1542
1543[!] Note that special memory barrier primitives are available for these
1544situations because on some CPUs the atomic instructions used imply full memory
1545barriers, and so barrier instructions are superfluous in conjunction with them,
1546and in such cases the special barrier primitives will be no-ops.
David Howells108b42b2006-03-31 16:00:29 +01001547
1548See Documentation/atomic_ops.txt for more information.
1549
1550
1551ACCESSING DEVICES
1552-----------------
1553
1554Many devices can be memory mapped, and so appear to the CPU as if they're just
1555a set of memory locations. To control such a device, the driver usually has to
1556make the right memory accesses in exactly the right order.
1557
1558However, having a clever CPU or a clever compiler creates a potential problem
1559in that the carefully sequenced accesses in the driver code won't reach the
1560device in the requisite order if the CPU or the compiler thinks it is more
1561efficient to reorder, combine or merge accesses - something that would cause
1562the device to malfunction.
1563
1564Inside of the Linux kernel, I/O should be done through the appropriate accessor
1565routines - such as inb() or writel() - which know how to make such accesses
1566appropriately sequential. Whilst this, for the most part, renders the explicit
1567use of memory barriers unnecessary, there are a couple of situations where they
1568might be needed:
1569
1570 (1) On some systems, I/O stores are not strongly ordered across all CPUs, and
1571 so for _all_ general drivers locks should be used and mmiowb() must be
1572 issued prior to unlocking the critical section.
1573
1574 (2) If the accessor functions are used to refer to an I/O memory window with
1575 relaxed memory access properties, then _mandatory_ memory barriers are
1576 required to enforce ordering.
1577
1578See Documentation/DocBook/deviceiobook.tmpl for more information.
1579
1580
1581INTERRUPTS
1582----------
1583
1584A driver may be interrupted by its own interrupt service routine, and thus the
1585two parts of the driver may interfere with each other's attempts to control or
1586access the device.
1587
1588This may be alleviated - at least in part - by disabling local interrupts (a
1589form of locking), such that the critical operations are all contained within
1590the interrupt-disabled section in the driver. Whilst the driver's interrupt
1591routine is executing, the driver's core may not run on the same CPU, and its
1592interrupt is not permitted to happen again until the current interrupt has been
1593handled, thus the interrupt handler does not need to lock against that.
1594
1595However, consider a driver that was talking to an ethernet card that sports an
1596address register and a data register. If that driver's core talks to the card
1597under interrupt-disablement and then the driver's interrupt handler is invoked:
1598
1599 LOCAL IRQ DISABLE
1600 writew(ADDR, 3);
1601 writew(DATA, y);
1602 LOCAL IRQ ENABLE
1603 <interrupt>
1604 writew(ADDR, 4);
1605 q = readw(DATA);
1606 </interrupt>
1607
1608The store to the data register might happen after the second store to the
1609address register if ordering rules are sufficiently relaxed:
1610
1611 STORE *ADDR = 3, STORE *ADDR = 4, STORE *DATA = y, q = LOAD *DATA
1612
1613
1614If ordering rules are relaxed, it must be assumed that accesses done inside an
1615interrupt disabled section may leak outside of it and may interleave with
1616accesses performed in an interrupt - and vice versa - unless implicit or
1617explicit barriers are used.
1618
1619Normally this won't be a problem because the I/O accesses done inside such
1620sections will include synchronous load operations on strictly ordered I/O
1621registers that form implicit I/O barriers. If this isn't sufficient then an
1622mmiowb() may need to be used explicitly.
1623
1624
1625A similar situation may occur between an interrupt routine and two routines
1626running on separate CPUs that communicate with each other. If such a case is
1627likely, then interrupt-disabling locks should be used to guarantee ordering.
1628
1629
1630==========================
1631KERNEL I/O BARRIER EFFECTS
1632==========================
1633
1634When accessing I/O memory, drivers should use the appropriate accessor
1635functions:
1636
1637 (*) inX(), outX():
1638
1639 These are intended to talk to I/O space rather than memory space, but
1640 that's primarily a CPU-specific concept. The i386 and x86_64 processors do
1641 indeed have special I/O space access cycles and instructions, but many
1642 CPUs don't have such a concept.
1643
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001644 The PCI bus, amongst others, defines an I/O space concept which - on such
1645 CPUs as i386 and x86_64 - readily maps to the CPU's concept of I/O
David Howells6bc39272006-06-25 05:49:22 -07001646 space. However, it may also be mapped as a virtual I/O space in the CPU's
1647 memory map, particularly on those CPUs that don't support alternate I/O
1648 spaces.
David Howells108b42b2006-03-31 16:00:29 +01001649
1650 Accesses to this space may be fully synchronous (as on i386), but
1651 intermediary bridges (such as the PCI host bridge) may not fully honour
1652 that.
1653
1654 They are guaranteed to be fully ordered with respect to each other.
1655
1656 They are not guaranteed to be fully ordered with respect to other types of
1657 memory and I/O operation.
1658
1659 (*) readX(), writeX():
1660
1661 Whether these are guaranteed to be fully ordered and uncombined with
1662 respect to each other on the issuing CPU depends on the characteristics
1663 defined for the memory window through which they're accessing. On later
1664 i386 architecture machines, for example, this is controlled by way of the
1665 MTRR registers.
1666
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001667 Ordinarily, these will be guaranteed to be fully ordered and uncombined,
David Howells108b42b2006-03-31 16:00:29 +01001668 provided they're not accessing a prefetchable device.
1669
1670 However, intermediary hardware (such as a PCI bridge) may indulge in
1671 deferral if it so wishes; to flush a store, a load from the same location
1672 is preferred[*], but a load from the same device or from configuration
1673 space should suffice for PCI.
1674
1675 [*] NOTE! attempting to load from the same location as was written to may
1676 cause a malfunction - consider the 16550 Rx/Tx serial registers for
1677 example.
1678
1679 Used with prefetchable I/O memory, an mmiowb() barrier may be required to
1680 force stores to be ordered.
1681
1682 Please refer to the PCI specification for more information on interactions
1683 between PCI transactions.
1684
1685 (*) readX_relaxed()
1686
1687 These are similar to readX(), but are not guaranteed to be ordered in any
1688 way. Be aware that there is no I/O read barrier available.
1689
1690 (*) ioreadX(), iowriteX()
1691
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001692 These will perform appropriately for the type of access they're actually
David Howells108b42b2006-03-31 16:00:29 +01001693 doing, be it inX()/outX() or readX()/writeX().
1694
1695
1696========================================
1697ASSUMED MINIMUM EXECUTION ORDERING MODEL
1698========================================
1699
1700It has to be assumed that the conceptual CPU is weakly-ordered but that it will
1701maintain the appearance of program causality with respect to itself. Some CPUs
1702(such as i386 or x86_64) are more constrained than others (such as powerpc or
1703frv), and so the most relaxed case (namely DEC Alpha) must be assumed outside
1704of arch-specific code.
1705
1706This means that it must be considered that the CPU will execute its instruction
1707stream in any order it feels like - or even in parallel - provided that if an
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001708instruction in the stream depends on an earlier instruction, then that
David Howells108b42b2006-03-31 16:00:29 +01001709earlier instruction must be sufficiently complete[*] before the later
1710instruction may proceed; in other words: provided that the appearance of
1711causality is maintained.
1712
1713 [*] Some instructions have more than one effect - such as changing the
1714 condition codes, changing registers or changing memory - and different
1715 instructions may depend on different effects.
1716
1717A CPU may also discard any instruction sequence that winds up having no
1718ultimate effect. For example, if two adjacent instructions both load an
1719immediate value into the same register, the first may be discarded.
1720
1721
1722Similarly, it has to be assumed that compiler might reorder the instruction
1723stream in any way it sees fit, again provided the appearance of causality is
1724maintained.
1725
1726
1727============================
1728THE EFFECTS OF THE CPU CACHE
1729============================
1730
1731The way cached memory operations are perceived across the system is affected to
1732a certain extent by the caches that lie between CPUs and memory, and by the
1733memory coherence system that maintains the consistency of state in the system.
1734
1735As far as the way a CPU interacts with another part of the system through the
1736caches goes, the memory system has to include the CPU's caches, and memory
1737barriers for the most part act at the interface between the CPU and its cache
1738(memory barriers logically act on the dotted line in the following diagram):
1739
1740 <--- CPU ---> : <----------- Memory ----------->
1741 :
1742 +--------+ +--------+ : +--------+ +-----------+
1743 | | | | : | | | | +--------+
1744 | CPU | | Memory | : | CPU | | | | |
1745 | Core |--->| Access |----->| Cache |<-->| | | |
1746 | | | Queue | : | | | |--->| Memory |
1747 | | | | : | | | | | |
1748 +--------+ +--------+ : +--------+ | | | |
1749 : | Cache | +--------+
1750 : | Coherency |
1751 : | Mechanism | +--------+
1752 +--------+ +--------+ : +--------+ | | | |
1753 | | | | : | | | | | |
1754 | CPU | | Memory | : | CPU | | |--->| Device |
1755 | Core |--->| Access |----->| Cache |<-->| | | |
1756 | | | Queue | : | | | | | |
1757 | | | | : | | | | +--------+
1758 +--------+ +--------+ : +--------+ +-----------+
1759 :
1760 :
1761
1762Although any particular load or store may not actually appear outside of the
1763CPU that issued it since it may have been satisfied within the CPU's own cache,
1764it will still appear as if the full memory access had taken place as far as the
1765other CPUs are concerned since the cache coherency mechanisms will migrate the
1766cacheline over to the accessing CPU and propagate the effects upon conflict.
1767
1768The CPU core may execute instructions in any order it deems fit, provided the
1769expected program causality appears to be maintained. Some of the instructions
1770generate load and store operations which then go into the queue of memory
1771accesses to be performed. The core may place these in the queue in any order
1772it wishes, and continue execution until it is forced to wait for an instruction
1773to complete.
1774
1775What memory barriers are concerned with is controlling the order in which
1776accesses cross from the CPU side of things to the memory side of things, and
1777the order in which the effects are perceived to happen by the other observers
1778in the system.
1779
1780[!] Memory barriers are _not_ needed within a given CPU, as CPUs always see
1781their own loads and stores as if they had happened in program order.
1782
1783[!] MMIO or other device accesses may bypass the cache system. This depends on
1784the properties of the memory window through which devices are accessed and/or
1785the use of any special device communication instructions the CPU may have.
1786
1787
1788CACHE COHERENCY
1789---------------
1790
1791Life isn't quite as simple as it may appear above, however: for while the
1792caches are expected to be coherent, there's no guarantee that that coherency
1793will be ordered. This means that whilst changes made on one CPU will
1794eventually become visible on all CPUs, there's no guarantee that they will
1795become apparent in the same order on those other CPUs.
1796
1797
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001798Consider dealing with a system that has a pair of CPUs (1 & 2), each of which
1799has a pair of parallel data caches (CPU 1 has A/B, and CPU 2 has C/D):
David Howells108b42b2006-03-31 16:00:29 +01001800
1801 :
1802 : +--------+
1803 : +---------+ | |
1804 +--------+ : +--->| Cache A |<------->| |
1805 | | : | +---------+ | |
1806 | CPU 1 |<---+ | |
1807 | | : | +---------+ | |
1808 +--------+ : +--->| Cache B |<------->| |
1809 : +---------+ | |
1810 : | Memory |
1811 : +---------+ | System |
1812 +--------+ : +--->| Cache C |<------->| |
1813 | | : | +---------+ | |
1814 | CPU 2 |<---+ | |
1815 | | : | +---------+ | |
1816 +--------+ : +--->| Cache D |<------->| |
1817 : +---------+ | |
1818 : +--------+
1819 :
1820
1821Imagine the system has the following properties:
1822
1823 (*) an odd-numbered cache line may be in cache A, cache C or it may still be
1824 resident in memory;
1825
1826 (*) an even-numbered cache line may be in cache B, cache D or it may still be
1827 resident in memory;
1828
1829 (*) whilst the CPU core is interrogating one cache, the other cache may be
1830 making use of the bus to access the rest of the system - perhaps to
1831 displace a dirty cacheline or to do a speculative load;
1832
1833 (*) each cache has a queue of operations that need to be applied to that cache
1834 to maintain coherency with the rest of the system;
1835
1836 (*) the coherency queue is not flushed by normal loads to lines already
1837 present in the cache, even though the contents of the queue may
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001838 potentially affect those loads.
David Howells108b42b2006-03-31 16:00:29 +01001839
1840Imagine, then, that two writes are made on the first CPU, with a write barrier
1841between them to guarantee that they will appear to reach that CPU's caches in
1842the requisite order:
1843
1844 CPU 1 CPU 2 COMMENT
1845 =============== =============== =======================================
1846 u == 0, v == 1 and p == &u, q == &u
1847 v = 2;
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001848 smp_wmb(); Make sure change to v is visible before
David Howells108b42b2006-03-31 16:00:29 +01001849 change to p
1850 <A:modify v=2> v is now in cache A exclusively
1851 p = &v;
1852 <B:modify p=&v> p is now in cache B exclusively
1853
1854The write memory barrier forces the other CPUs in the system to perceive that
1855the local CPU's caches have apparently been updated in the correct order. But
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001856now imagine that the second CPU wants to read those values:
David Howells108b42b2006-03-31 16:00:29 +01001857
1858 CPU 1 CPU 2 COMMENT
1859 =============== =============== =======================================
1860 ...
1861 q = p;
1862 x = *q;
1863
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001864The above pair of reads may then fail to happen in the expected order, as the
David Howells108b42b2006-03-31 16:00:29 +01001865cacheline holding p may get updated in one of the second CPU's caches whilst
1866the update to the cacheline holding v is delayed in the other of the second
1867CPU's caches by some other cache event:
1868
1869 CPU 1 CPU 2 COMMENT
1870 =============== =============== =======================================
1871 u == 0, v == 1 and p == &u, q == &u
1872 v = 2;
1873 smp_wmb();
1874 <A:modify v=2> <C:busy>
1875 <C:queue v=2>
Aneesh Kumar79afecf2006-05-15 09:44:36 -07001876 p = &v; q = p;
David Howells108b42b2006-03-31 16:00:29 +01001877 <D:request p>
1878 <B:modify p=&v> <D:commit p=&v>
1879 <D:read p>
1880 x = *q;
1881 <C:read *q> Reads from v before v updated in cache
1882 <C:unbusy>
1883 <C:commit v=2>
1884
1885Basically, whilst both cachelines will be updated on CPU 2 eventually, there's
1886no guarantee that, without intervention, the order of update will be the same
1887as that committed on CPU 1.
1888
1889
1890To intervene, we need to interpolate a data dependency barrier or a read
1891barrier between the loads. This will force the cache to commit its coherency
1892queue before processing any further requests:
1893
1894 CPU 1 CPU 2 COMMENT
1895 =============== =============== =======================================
1896 u == 0, v == 1 and p == &u, q == &u
1897 v = 2;
1898 smp_wmb();
1899 <A:modify v=2> <C:busy>
1900 <C:queue v=2>
Paolo 'Blaisorblade' Giarrusso3fda9822006-10-19 23:28:19 -07001901 p = &v; q = p;
David Howells108b42b2006-03-31 16:00:29 +01001902 <D:request p>
1903 <B:modify p=&v> <D:commit p=&v>
1904 <D:read p>
1905 smp_read_barrier_depends()
1906 <C:unbusy>
1907 <C:commit v=2>
1908 x = *q;
1909 <C:read *q> Reads from v after v updated in cache
1910
1911
1912This sort of problem can be encountered on DEC Alpha processors as they have a
1913split cache that improves performance by making better use of the data bus.
1914Whilst most CPUs do imply a data dependency barrier on the read when a memory
1915access depends on a read, not all do, so it may not be relied on.
1916
1917Other CPUs may also have split caches, but must coordinate between the various
Matt LaPlante3f6dee92006-10-03 22:45:33 +02001918cachelets for normal memory accesses. The semantics of the Alpha removes the
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001919need for coordination in the absence of memory barriers.
David Howells108b42b2006-03-31 16:00:29 +01001920
1921
1922CACHE COHERENCY VS DMA
1923----------------------
1924
1925Not all systems maintain cache coherency with respect to devices doing DMA. In
1926such cases, a device attempting DMA may obtain stale data from RAM because
1927dirty cache lines may be resident in the caches of various CPUs, and may not
1928have been written back to RAM yet. To deal with this, the appropriate part of
1929the kernel must flush the overlapping bits of cache on each CPU (and maybe
1930invalidate them as well).
1931
1932In addition, the data DMA'd to RAM by a device may be overwritten by dirty
1933cache lines being written back to RAM from a CPU's cache after the device has
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001934installed its own data, or cache lines present in the CPU's cache may simply
1935obscure the fact that RAM has been updated, until at such time as the cacheline
1936is discarded from the CPU's cache and reloaded. To deal with this, the
1937appropriate part of the kernel must invalidate the overlapping bits of the
David Howells108b42b2006-03-31 16:00:29 +01001938cache on each CPU.
1939
1940See Documentation/cachetlb.txt for more information on cache management.
1941
1942
1943CACHE COHERENCY VS MMIO
1944-----------------------
1945
1946Memory mapped I/O usually takes place through memory locations that are part of
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001947a window in the CPU's memory space that has different properties assigned than
David Howells108b42b2006-03-31 16:00:29 +01001948the usual RAM directed window.
1949
1950Amongst these properties is usually the fact that such accesses bypass the
1951caching entirely and go directly to the device buses. This means MMIO accesses
1952may, in effect, overtake accesses to cached memory that were emitted earlier.
1953A memory barrier isn't sufficient in such a case, but rather the cache must be
1954flushed between the cached memory write and the MMIO access if the two are in
1955any way dependent.
1956
1957
1958=========================
1959THE THINGS CPUS GET UP TO
1960=========================
1961
1962A programmer might take it for granted that the CPU will perform memory
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001963operations in exactly the order specified, so that if the CPU is, for example,
David Howells108b42b2006-03-31 16:00:29 +01001964given the following piece of code to execute:
1965
1966 a = *A;
1967 *B = b;
1968 c = *C;
1969 d = *D;
1970 *E = e;
1971
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001972they would then expect that the CPU will complete the memory operation for each
David Howells108b42b2006-03-31 16:00:29 +01001973instruction before moving on to the next one, leading to a definite sequence of
1974operations as seen by external observers in the system:
1975
1976 LOAD *A, STORE *B, LOAD *C, LOAD *D, STORE *E.
1977
1978
1979Reality is, of course, much messier. With many CPUs and compilers, the above
1980assumption doesn't hold because:
1981
1982 (*) loads are more likely to need to be completed immediately to permit
1983 execution progress, whereas stores can often be deferred without a
1984 problem;
1985
1986 (*) loads may be done speculatively, and the result discarded should it prove
1987 to have been unnecessary;
1988
Jarek Poplawski81fc6322007-05-23 13:58:20 -07001989 (*) loads may be done speculatively, leading to the result having been fetched
1990 at the wrong time in the expected sequence of events;
David Howells108b42b2006-03-31 16:00:29 +01001991
1992 (*) the order of the memory accesses may be rearranged to promote better use
1993 of the CPU buses and caches;
1994
1995 (*) loads and stores may be combined to improve performance when talking to
1996 memory or I/O hardware that can do batched accesses of adjacent locations,
1997 thus cutting down on transaction setup costs (memory and PCI devices may
1998 both be able to do this); and
1999
2000 (*) the CPU's data cache may affect the ordering, and whilst cache-coherency
2001 mechanisms may alleviate this - once the store has actually hit the cache
2002 - there's no guarantee that the coherency management will be propagated in
2003 order to other CPUs.
2004
2005So what another CPU, say, might actually observe from the above piece of code
2006is:
2007
2008 LOAD *A, ..., LOAD {*C,*D}, STORE *E, STORE *B
2009
2010 (Where "LOAD {*C,*D}" is a combined load)
2011
2012
2013However, it is guaranteed that a CPU will be self-consistent: it will see its
2014_own_ accesses appear to be correctly ordered, without the need for a memory
2015barrier. For instance with the following code:
2016
2017 U = *A;
2018 *A = V;
2019 *A = W;
2020 X = *A;
2021 *A = Y;
2022 Z = *A;
2023
2024and assuming no intervention by an external influence, it can be assumed that
2025the final result will appear to be:
2026
2027 U == the original value of *A
2028 X == W
2029 Z == Y
2030 *A == Y
2031
2032The code above may cause the CPU to generate the full sequence of memory
2033accesses:
2034
2035 U=LOAD *A, STORE *A=V, STORE *A=W, X=LOAD *A, STORE *A=Y, Z=LOAD *A
2036
2037in that order, but, without intervention, the sequence may have almost any
2038combination of elements combined or discarded, provided the program's view of
2039the world remains consistent.
2040
2041The compiler may also combine, discard or defer elements of the sequence before
2042the CPU even sees them.
2043
2044For instance:
2045
2046 *A = V;
2047 *A = W;
2048
2049may be reduced to:
2050
2051 *A = W;
2052
2053since, without a write barrier, it can be assumed that the effect of the
2054storage of V to *A is lost. Similarly:
2055
2056 *A = Y;
2057 Z = *A;
2058
2059may, without a memory barrier, be reduced to:
2060
2061 *A = Y;
2062 Z = Y;
2063
2064and the LOAD operation never appear outside of the CPU.
2065
2066
2067AND THEN THERE'S THE ALPHA
2068--------------------------
2069
2070The DEC Alpha CPU is one of the most relaxed CPUs there is. Not only that,
2071some versions of the Alpha CPU have a split data cache, permitting them to have
Jarek Poplawski81fc6322007-05-23 13:58:20 -07002072two semantically-related cache lines updated at separate times. This is where
David Howells108b42b2006-03-31 16:00:29 +01002073the data dependency barrier really becomes necessary as this synchronises both
2074caches with the memory coherence system, thus making it seem like pointer
2075changes vs new data occur in the right order.
2076
Jarek Poplawski81fc6322007-05-23 13:58:20 -07002077The Alpha defines the Linux kernel's memory barrier model.
David Howells108b42b2006-03-31 16:00:29 +01002078
2079See the subsection on "Cache Coherency" above.
2080
2081
2082==========
2083REFERENCES
2084==========
2085
2086Alpha AXP Architecture Reference Manual, Second Edition (Sites & Witek,
2087Digital Press)
2088 Chapter 5.2: Physical Address Space Characteristics
2089 Chapter 5.4: Caches and Write Buffers
2090 Chapter 5.5: Data Sharing
2091 Chapter 5.6: Read/Write Ordering
2092
2093AMD64 Architecture Programmer's Manual Volume 2: System Programming
2094 Chapter 7.1: Memory-Access Ordering
2095 Chapter 7.4: Buffering and Combining Memory Writes
2096
2097IA-32 Intel Architecture Software Developer's Manual, Volume 3:
2098System Programming Guide
2099 Chapter 7.1: Locked Atomic Operations
2100 Chapter 7.2: Memory Ordering
2101 Chapter 7.4: Serializing Instructions
2102
2103The SPARC Architecture Manual, Version 9
2104 Chapter 8: Memory Models
2105 Appendix D: Formal Specification of the Memory Models
2106 Appendix J: Programming with the Memory Models
2107
2108UltraSPARC Programmer Reference Manual
2109 Chapter 5: Memory Accesses and Cacheability
2110 Chapter 15: Sparc-V9 Memory Models
2111
2112UltraSPARC III Cu User's Manual
2113 Chapter 9: Memory Models
2114
2115UltraSPARC IIIi Processor User's Manual
2116 Chapter 8: Memory Models
2117
2118UltraSPARC Architecture 2005
2119 Chapter 9: Memory
2120 Appendix D: Formal Specifications of the Memory Models
2121
2122UltraSPARC T1 Supplement to the UltraSPARC Architecture 2005
2123 Chapter 8: Memory Models
2124 Appendix F: Caches and Cache Coherency
2125
2126Solaris Internals, Core Kernel Architecture, p63-68:
2127 Chapter 3.3: Hardware Considerations for Locks and
2128 Synchronization
2129
2130Unix Systems for Modern Architectures, Symmetric Multiprocessing and Caching
2131for Kernel Programmers:
2132 Chapter 13: Other Memory Models
2133
2134Intel Itanium Architecture Software Developer's Manual: Volume 1:
2135 Section 2.6: Speculation
2136 Section 4.4: Memory Access