| blob | 0d8d7ef131e9aa1782c1f591624723e483a5bd04 |
1 ============================
2 LINUX KERNEL MEMORY BARRIERS
3 ============================
5 By: David Howells <dhowells@redhat.com>
6 Paul E. McKenney <paulmck@linux.vnet.ibm.com>
7 Will Deacon <will.deacon@arm.com>
8 Peter Zijlstra <peterz@infradead.org>
10 ==========
11 DISCLAIMER
12 ==========
14 This document is not a specification; it is intentionally (for the sake of
15 brevity) and unintentionally (due to being human) incomplete. This document is
16 meant as a guide to using the various memory barriers provided by Linux, but
17 in case of any doubt (and there are many) please ask. Some doubts may be
18 resolved by referring to the formal memory consistency model and related
19 documentation at tools/memory-model/. Nevertheless, even this memory
20 model should be viewed as the collective opinion of its maintainers rather
21 than as an infallible oracle.
23 To repeat, this document is not a specification of what Linux expects from
24 hardware.
26 The purpose of this document is twofold:
28 (1) to specify the minimum functionality that one can rely on for any
29 particular barrier, and
31 (2) to provide a guide as to how to use the barriers that are available.
33 Note that an architecture can provide more than the minimum requirement
34 for any particular barrier, but if the architecture provides less than
35 that, that architecture is incorrect.
37 Note also that it is possible that a barrier may be a no-op for an
38 architecture because the way that arch works renders an explicit barrier
39 unnecessary in that case.
42 ========
43 CONTENTS
44 ========
46 (*) Abstract memory access model.
48 - Device operations.
49 - Guarantees.
51 (*) What are memory barriers?
53 - Varieties of memory barrier.
54 - What may not be assumed about memory barriers?
55 - Data dependency barriers (historical).
56 - Control dependencies.
57 - SMP barrier pairing.
58 - Examples of memory barrier sequences.
59 - Read memory barriers vs load speculation.
60 - Multicopy atomicity.
62 (*) Explicit kernel barriers.
64 - Compiler barrier.
65 - CPU memory barriers.
66 - MMIO write barrier.
68 (*) Implicit kernel memory barriers.
70 - Lock acquisition functions.
71 - Interrupt disabling functions.
72 - Sleep and wake-up functions.
73 - Miscellaneous functions.
75 (*) Inter-CPU acquiring barrier effects.
77 - Acquires vs memory accesses.
78 - Acquires vs I/O accesses.
80 (*) Where are memory barriers needed?
82 - Interprocessor interaction.
83 - Atomic operations.
84 - Accessing devices.
85 - Interrupts.
87 (*) Kernel I/O barrier effects.
89 (*) Assumed minimum execution ordering model.
91 (*) The effects of the cpu cache.
93 - Cache coherency.
94 - Cache coherency vs DMA.
95 - Cache coherency vs MMIO.
97 (*) The things CPUs get up to.
99 - And then there's the Alpha.
100 - Virtual Machine Guests.
102 (*) Example uses.
104 - Circular buffers.
106 (*) References.
109 ============================
110 ABSTRACT MEMORY ACCESS MODEL
111 ============================
113 Consider the following abstract model of the system:
115 : :
116 : :
117 : :
118 +-------+ : +--------+ : +-------+
119 | | : | | : | |
120 | | : | | : | |
121 | CPU 1 |<----->| Memory |<----->| CPU 2 |
122 | | : | | : | |
123 | | : | | : | |
124 +-------+ : +--------+ : +-------+
125 ^ : ^ : ^
126 | : | : |
127 | : | : |
128 | : v : |
129 | : +--------+ : |
130 | : | | : |
131 | : | | : |
132 +---------->| Device |<----------+
133 : | | :
134 : | | :
135 : +--------+ :
136 : :
138 Each CPU executes a program that generates memory access operations. In the
139 abstract CPU, memory operation ordering is very relaxed, and a CPU may actually
140 perform the memory operations in any order it likes, provided program causality
141 appears to be maintained. Similarly, the compiler may also arrange the
142 instructions it emits in any order it likes, provided it doesn't affect the
143 apparent operation of the program.
145 So in the above diagram, the effects of the memory operations performed by a
146 CPU are perceived by the rest of the system as the operations cross the
147 interface between the CPU and rest of the system (the dotted lines).
150 For example, consider the following sequence of events:
152 CPU 1 CPU 2
153 =============== ===============
154 { A == 1; B == 2 }
155 A = 3; x = B;
156 B = 4; y = A;
158 The set of accesses as seen by the memory system in the middle can be arranged
159 in 24 different combinations:
161 STORE A=3, STORE B=4, y=LOAD A->3, x=LOAD B->4
162 STORE A=3, STORE B=4, x=LOAD B->4, y=LOAD A->3
163 STORE A=3, y=LOAD A->3, STORE B=4, x=LOAD B->4
164 STORE A=3, y=LOAD A->3, x=LOAD B->2, STORE B=4
165 STORE A=3, x=LOAD B->2, STORE B=4, y=LOAD A->3
166 STORE A=3, x=LOAD B->2, y=LOAD A->3, STORE B=4
167 STORE B=4, STORE A=3, y=LOAD A->3, x=LOAD B->4
168 STORE B=4, ...
169 ...
171 and can thus result in four different combinations of values:
173 x == 2, y == 1
174 x == 2, y == 3
175 x == 4, y == 1
176 x == 4, y == 3
179 Furthermore, the stores committed by a CPU to the memory system may not be
180 perceived by the loads made by another CPU in the same order as the stores were
181 committed.
184 As a further example, consider this sequence of events:
186 CPU 1 CPU 2
187 =============== ===============
188 { A == 1, B == 2, C == 3, P == &A, Q == &C }
189 B = 4; Q = P;
190 P = &B D = *Q;
192 There is an obvious data dependency here, as the value loaded into D depends on
193 the address retrieved from P by CPU 2. At the end of the sequence, any of the
194 following results are possible:
196 (Q == &A) and (D == 1)
197 (Q == &B) and (D == 2)
198 (Q == &B) and (D == 4)
200 Note that CPU 2 will never try and load C into D because the CPU will load P
201 into Q before issuing the load of *Q.
204 DEVICE OPERATIONS
205 -----------------
207 Some devices present their control interfaces as collections of memory
208 locations, but the order in which the control registers are accessed is very
209 important. For instance, imagine an ethernet card with a set of internal
210 registers that are accessed through an address port register (A) and a data
211 port register (D). To read internal register 5, the following code might then
212 be used:
214 *A = 5;
215 x = *D;
217 but this might show up as either of the following two sequences:
219 STORE *A = 5, x = LOAD *D
220 x = LOAD *D, STORE *A = 5
222 the second of which will almost certainly result in a malfunction, since it set
223 the address _after_ attempting to read the register.
226 GUARANTEES
227 ----------
229 There are some minimal guarantees that may be expected of a CPU:
231 (*) On any given CPU, dependent memory accesses will be issued in order, with
232 respect to itself. This means that for:
234 Q = READ_ONCE(P); D = READ_ONCE(*Q);
236 the CPU will issue the following memory operations:
238 Q = LOAD P, D = LOAD *Q
240 and always in that order. However, on DEC Alpha, READ_ONCE() also
241 emits a memory-barrier instruction, so that a DEC Alpha CPU will
242 instead issue the following memory operations:
244 Q = LOAD P, MEMORY_BARRIER, D = LOAD *Q, MEMORY_BARRIER
246 Whether on DEC Alpha or not, the READ_ONCE() also prevents compiler
247 mischief.
249 (*) Overlapping loads and stores within a particular CPU will appear to be
250 ordered within that CPU. This means that for:
252 a = READ_ONCE(*X); WRITE_ONCE(*X, b);
254 the CPU will only issue the following sequence of memory operations:
256 a = LOAD *X, STORE *X = b
258 And for:
260 WRITE_ONCE(*X, c); d = READ_ONCE(*X);
262 the CPU will only issue:
264 STORE *X = c, d = LOAD *X
266 (Loads and stores overlap if they are targeted at overlapping pieces of
267 memory).
269 And there are a number of things that _must_ or _must_not_ be assumed:
271 (*) It _must_not_ be assumed that the compiler will do what you want
272 with memory references that are not protected by READ_ONCE() and
273 WRITE_ONCE(). Without them, the compiler is within its rights to
274 do all sorts of "creative" transformations, which are covered in
275 the COMPILER BARRIER section.
277 (*) It _must_not_ be assumed that independent loads and stores will be issued
278 in the order given. This means that for:
280 X = *A; Y = *B; *D = Z;
282 we may get any of the following sequences:
284 X = LOAD *A, Y = LOAD *B, STORE *D = Z
285 X = LOAD *A, STORE *D = Z, Y = LOAD *B
286 Y = LOAD *B, X = LOAD *A, STORE *D = Z
287 Y = LOAD *B, STORE *D = Z, X = LOAD *A
288 STORE *D = Z, X = LOAD *A, Y = LOAD *B
289 STORE *D = Z, Y = LOAD *B, X = LOAD *A
291 (*) It _must_ be assumed that overlapping memory accesses may be merged or
292 discarded. This means that for:
294 X = *A; Y = *(A + 4);
296 we may get any one of the following sequences:
298 X = LOAD *A; Y = LOAD *(A + 4);
299 Y = LOAD *(A + 4); X = LOAD *A;
300 {X, Y} = LOAD {*A, *(A + 4) };
302 And for:
304 *A = X; *(A + 4) = Y;
306 we may get any of:
308 STORE *A = X; STORE *(A + 4) = Y;
309 STORE *(A + 4) = Y; STORE *A = X;
310 STORE {*A, *(A + 4) } = {X, Y};
312 And there are anti-guarantees:
314 (*) These guarantees do not apply to bitfields, because compilers often
315 generate code to modify these using non-atomic read-modify-write
316 sequences. Do not attempt to use bitfields to synchronize parallel
317 algorithms.
319 (*) Even in cases where bitfields are protected by locks, all fields
320 in a given bitfield must be protected by one lock. If two fields
321 in a given bitfield are protected by different locks, the compiler's
322 non-atomic read-modify-write sequences can cause an update to one
323 field to corrupt the value of an adjacent field.
325 (*) These guarantees apply only to properly aligned and sized scalar
326 variables. "Properly sized" currently means variables that are
327 the same size as "char", "short", "int" and "long". "Properly
328 aligned" means the natural alignment, thus no constraints for
329 "char", two-byte alignment for "short", four-byte alignment for
330 "int", and either four-byte or eight-byte alignment for "long",
331 on 32-bit and 64-bit systems, respectively. Note that these
332 guarantees were introduced into the C11 standard, so beware when
333 using older pre-C11 compilers (for example, gcc 4.6). The portion
334 of the standard containing this guarantee is Section 3.14, which
335 defines "memory location" as follows:
337 memory location
338 either an object of scalar type, or a maximal sequence
339 of adjacent bit-fields all having nonzero width
341 NOTE 1: Two threads of execution can update and access
342 separate memory locations without interfering with
343 each other.
345 NOTE 2: A bit-field and an adjacent non-bit-field member
346 are in separate memory locations. The same applies
347 to two bit-fields, if one is declared inside a nested
348 structure declaration and the other is not, or if the two
349 are separated by a zero-length bit-field declaration,
350 or if they are separated by a non-bit-field member
351 declaration. It is not safe to concurrently update two
352 bit-fields in the same structure if all members declared
353 between them are also bit-fields, no matter what the
354 sizes of those intervening bit-fields happen to be.
357 =========================
358 WHAT ARE MEMORY BARRIERS?
359 =========================
361 As can be seen above, independent memory operations are effectively performed
362 in random order, but this can be a problem for CPU-CPU interaction and for I/O.
363 What is required is some way of intervening to instruct the compiler and the
364 CPU to restrict the order.
366 Memory barriers are such interventions. They impose a perceived partial
367 ordering over the memory operations on either side of the barrier.
369 Such enforcement is important because the CPUs and other devices in a system
370 can use a variety of tricks to improve performance, including reordering,
371 deferral and combination of memory operations; speculative loads; speculative
372 branch prediction and various types of caching. Memory barriers are used to
373 override or suppress these tricks, allowing the code to sanely control the
374 interaction of multiple CPUs and/or devices.
377 VARIETIES OF MEMORY BARRIER
378 ---------------------------
380 Memory barriers come in four basic varieties:
382 (1) Write (or store) memory barriers.
384 A write memory barrier gives a guarantee that all the STORE operations
385 specified before the barrier will appear to happen before all the STORE
386 operations specified after the barrier with respect to the other
387 components of the system.
389 A write barrier is a partial ordering on stores only; it is not required
390 to have any effect on loads.
392 A CPU can be viewed as committing a sequence of store operations to the
393 memory system as time progresses. All stores _before_ a write barrier
394 will occur _before_ all the stores after the write barrier.
396 [!] Note that write barriers should normally be paired with read or data
397 dependency barriers; see the "SMP barrier pairing" subsection.
400 (2) Data dependency barriers.
402 A data dependency barrier is a weaker form of read barrier. In the case
403 where two loads are performed such that the second depends on the result
404 of the first (eg: the first load retrieves the address to which the second
405 load will be directed), a data dependency barrier would be required to
406 make sure that the target of the second load is updated after the address
407 obtained by the first load is accessed.
409 A data dependency barrier is a partial ordering on interdependent loads
410 only; it is not required to have any effect on stores, independent loads
411 or overlapping loads.
413 As mentioned in (1), the other CPUs in the system can be viewed as
414 committing sequences of stores to the memory system that the CPU being
415 considered can then perceive. A data dependency barrier issued by the CPU
416 under consideration guarantees that for any load preceding it, if that
417 load touches one of a sequence of stores from another CPU, then by the
418 time the barrier completes, the effects of all the stores prior to that
419 touched by the load will be perceptible to any loads issued after the data
420 dependency barrier.
422 See the "Examples of memory barrier sequences" subsection for diagrams
423 showing the ordering constraints.
425 [!] Note that the first load really has to have a _data_ dependency and
426 not a control dependency. If the address for the second load is dependent
427 on the first load, but the dependency is through a conditional rather than
428 actually loading the address itself, then it's a _control_ dependency and
429 a full read barrier or better is required. See the "Control dependencies"
430 subsection for more information.
432 [!] Note that data dependency barriers should normally be paired with
433 write barriers; see the "SMP barrier pairing" subsection.
436 (3) Read (or load) memory barriers.
438 A read barrier is a data dependency barrier plus a guarantee that all the
439 LOAD operations specified before the barrier will appear to happen before
440 all the LOAD operations specified after the barrier with respect to the
441 other components of the system.
443 A read barrier is a partial ordering on loads only; it is not required to
444 have any effect on stores.
446 Read memory barriers imply data dependency barriers, and so can substitute
447 for them.
449 [!] Note that read barriers should normally be paired with write barriers;
450 see the "SMP barrier pairing" subsection.
453 (4) General memory barriers.
455 A general memory barrier gives a guarantee that all the LOAD and STORE
456 operations specified before the barrier will appear to happen before all
457 the LOAD and STORE operations specified after the barrier with respect to
458 the other components of the system.
460 A general memory barrier is a partial ordering over both loads and stores.
462 General memory barriers imply both read and write memory barriers, and so
463 can substitute for either.
466 And a couple of implicit varieties:
468 (5) ACQUIRE operations.
470 This acts as a one-way permeable barrier. It guarantees that all memory
471 operations after the ACQUIRE operation will appear to happen after the
472 ACQUIRE operation with respect to the other components of the system.
473 ACQUIRE operations include LOCK operations and both smp_load_acquire()
474 and smp_cond_acquire() operations. The later builds the necessary ACQUIRE
475 semantics from relying on a control dependency and smp_rmb().
477 Memory operations that occur before an ACQUIRE operation may appear to
478 happen after it completes.
480 An ACQUIRE operation should almost always be paired with a RELEASE
481 operation.
484 (6) RELEASE operations.
486 This also acts as a one-way permeable barrier. It guarantees that all
487 memory operations before the RELEASE operation will appear to happen
488 before the RELEASE operation with respect to the other components of the
489 system. RELEASE operations include UNLOCK operations and
490 smp_store_release() operations.
492 Memory operations that occur after a RELEASE operation may appear to
493 happen before it completes.
495 The use of ACQUIRE and RELEASE operations generally precludes the need
496 for other sorts of memory barrier (but note the exceptions mentioned in
497 the subsection "MMIO write barrier"). In addition, a RELEASE+ACQUIRE
498 pair is -not- guaranteed to act as a full memory barrier. However, after
499 an ACQUIRE on a given variable, all memory accesses preceding any prior
500 RELEASE on that same variable are guaranteed to be visible. In other
501 words, within a given variable's critical section, all accesses of all
502 previous critical sections for that variable are guaranteed to have
503 completed.
505 This means that ACQUIRE acts as a minimal "acquire" operation and
506 RELEASE acts as a minimal "release" operation.
508 A subset of the atomic operations described in atomic_t.txt have ACQUIRE and
509 RELEASE variants in addition to fully-ordered and relaxed (no barrier
510 semantics) definitions. For compound atomics performing both a load and a
511 store, ACQUIRE semantics apply only to the load and RELEASE semantics apply
512 only to the store portion of the operation.
514 Memory barriers are only required where there's a possibility of interaction
515 between two CPUs or between a CPU and a device. If it can be guaranteed that
516 there won't be any such interaction in any particular piece of code, then
517 memory barriers are unnecessary in that piece of code.
520 Note that these are the _minimum_ guarantees. Different architectures may give
521 more substantial guarantees, but they may _not_ be relied upon outside of arch
522 specific code.
525 WHAT MAY NOT BE ASSUMED ABOUT MEMORY BARRIERS?
526 ----------------------------------------------
528 There are certain things that the Linux kernel memory barriers do not guarantee:
530 (*) There is no guarantee that any of the memory accesses specified before a
531 memory barrier will be _complete_ by the completion of a memory barrier
532 instruction; the barrier can be considered to draw a line in that CPU's
533 access queue that accesses of the appropriate type may not cross.
535 (*) There is no guarantee that issuing a memory barrier on one CPU will have
536 any direct effect on another CPU or any other hardware in the system. The
537 indirect effect will be the order in which the second CPU sees the effects
538 of the first CPU's accesses occur, but see the next point:
540 (*) There is no guarantee that a CPU will see the correct order of effects
541 from a second CPU's accesses, even _if_ the second CPU uses a memory
542 barrier, unless the first CPU _also_ uses a matching memory barrier (see
543 the subsection on "SMP Barrier Pairing").
545 (*) There is no guarantee that some intervening piece of off-the-CPU
546 hardware[*] will not reorder the memory accesses. CPU cache coherency
547 mechanisms should propagate the indirect effects of a memory barrier
548 between CPUs, but might not do so in order.
550 [*] For information on bus mastering DMA and coherency please read:
552 Documentation/PCI/pci.txt
553 Documentation/DMA-API-HOWTO.txt
554 Documentation/DMA-API.txt
557 DATA DEPENDENCY BARRIERS (HISTORICAL)
558 -------------------------------------
560 As of v4.15 of the Linux kernel, an smp_read_barrier_depends() was
561 added to READ_ONCE(), which means that about the only people who
562 need to pay attention to this section are those working on DEC Alpha
563 architecture-specific code and those working on READ_ONCE() itself.
564 For those who need it, and for those who are interested in the history,
565 here is the story of data-dependency barriers.
567 The usage requirements of data dependency barriers are a little subtle, and
568 it's not always obvious that they're needed. To illustrate, consider the
569 following sequence of events:
571 CPU 1 CPU 2
572 =============== ===============
573 { A == 1, B == 2, C == 3, P == &A, Q == &C }
574 B = 4;
575 <write barrier>
576 WRITE_ONCE(P, &B)
577 Q = READ_ONCE(P);
578 D = *Q;
580 There's a clear data dependency here, and it would seem that by the end of the
581 sequence, Q must be either &A or &B, and that:
583 (Q == &A) implies (D == 1)
584 (Q == &B) implies (D == 4)
586 But! CPU 2's perception of P may be updated _before_ its perception of B, thus
587 leading to the following situation:
589 (Q == &B) and (D == 2) ????
591 Whilst this may seem like a failure of coherency or causality maintenance, it
592 isn't, and this behaviour can be observed on certain real CPUs (such as the DEC
593 Alpha).
595 To deal with this, a data dependency barrier or better must be inserted
596 between the address load and the data load:
598 CPU 1 CPU 2
599 =============== ===============
600 { A == 1, B == 2, C == 3, P == &A, Q == &C }
601 B = 4;
602 <write barrier>
603 WRITE_ONCE(P, &B);
604 Q = READ_ONCE(P);
605 <data dependency barrier>
606 D = *Q;
608 This enforces the occurrence of one of the two implications, and prevents the
609 third possibility from arising.
612 [!] Note that this extremely counterintuitive situation arises most easily on
613 machines with split caches, so that, for example, one cache bank processes
614 even-numbered cache lines and the other bank processes odd-numbered cache
615 lines. The pointer P might be stored in an odd-numbered cache line, and the
616 variable B might be stored in an even-numbered cache line. Then, if the
617 even-numbered bank of the reading CPU's cache is extremely busy while the
618 odd-numbered bank is idle, one can see the new value of the pointer P (&B),
619 but the old value of the variable B (2).
622 A data-dependency barrier is not required to order dependent writes
623 because the CPUs that the Linux kernel supports don't do writes
624 until they are certain (1) that the write will actually happen, (2)
625 of the location of the write, and (3) of the value to be written.
626 But please carefully read the "CONTROL DEPENDENCIES" section and the
627 Documentation/RCU/rcu_dereference.txt file: The compiler can and does
628 break dependencies in a great many highly creative ways.
630 CPU 1 CPU 2
631 =============== ===============
632 { A == 1, B == 2, C = 3, P == &A, Q == &C }
633 B = 4;
634 <write barrier>
635 WRITE_ONCE(P, &B);
636 Q = READ_ONCE(P);
637 WRITE_ONCE(*Q, 5);
639 Therefore, no data-dependency barrier is required to order the read into
640 Q with the store into *Q. In other words, this outcome is prohibited,
641 even without a data-dependency barrier:
643 (Q == &B) && (B == 4)
645 Please note that this pattern should be rare. After all, the whole point
646 of dependency ordering is to -prevent- writes to the data structure, along
647 with the expensive cache misses associated with those writes. This pattern
648 can be used to record rare error conditions and the like, and the CPUs'
649 naturally occurring ordering prevents such records from being lost.
652 Note well that the ordering provided by a data dependency is local to
653 the CPU containing it. See the section on "Multicopy atomicity" for
654 more information.
657 The data dependency barrier is very important to the RCU system,
658 for example. See rcu_assign_pointer() and rcu_dereference() in
659 include/linux/rcupdate.h. This permits the current target of an RCU'd
660 pointer to be replaced with a new modified target, without the replacement
661 target appearing to be incompletely initialised.
663 See also the subsection on "Cache Coherency" for a more thorough example.
666 CONTROL DEPENDENCIES
667 --------------------
669 Control dependencies can be a bit tricky because current compilers do
670 not understand them. The purpose of this section is to help you prevent
671 the compiler's ignorance from breaking your code.
673 A load-load control dependency requires a full read memory barrier, not
674 simply a data dependency barrier to make it work correctly. Consider the
675 following bit of code:
677 q = READ_ONCE(a);
678 if (q) {
679 <data dependency barrier> /* BUG: No data dependency!!! */
680 p = READ_ONCE(b);
681 }
683 This will not have the desired effect because there is no actual data
684 dependency, but rather a control dependency that the CPU may short-circuit
685 by attempting to predict the outcome in advance, so that other CPUs see
686 the load from b as having happened before the load from a. In such a
687 case what's actually required is:
689 q = READ_ONCE(a);
690 if (q) {
691 <read barrier>
692 p = READ_ONCE(b);
693 }
695 However, stores are not speculated. This means that ordering -is- provided
696 for load-store control dependencies, as in the following example:
698 q = READ_ONCE(a);
699 if (q) {
700 WRITE_ONCE(b, 1);
701 }
703 Control dependencies pair normally with other types of barriers.
704 That said, please note that neither READ_ONCE() nor WRITE_ONCE()
705 are optional! Without the READ_ONCE(), the compiler might combine the
706 load from 'a' with other loads from 'a'. Without the WRITE_ONCE(),
707 the compiler might combine the store to 'b' with other stores to 'b'.
708 Either can result in highly counterintuitive effects on ordering.
710 Worse yet, if the compiler is able to prove (say) that the value of
711 variable 'a' is always non-zero, it would be well within its rights
712 to optimize the original example by eliminating the "if" statement
713 as follows:
715 q = a;
716 b = 1; /* BUG: Compiler and CPU can both reorder!!! */
718 So don't leave out the READ_ONCE().
720 It is tempting to try to enforce ordering on identical stores on both
721 branches of the "if" statement as follows:
723 q = READ_ONCE(a);
724 if (q) {
725 barrier();
726 WRITE_ONCE(b, 1);
727 do_something();
728 } else {
729 barrier();
730 WRITE_ONCE(b, 1);
731 do_something_else();
732 }
734 Unfortunately, current compilers will transform this as follows at high
735 optimization levels:
737 q = READ_ONCE(a);
738 barrier();
739 WRITE_ONCE(b, 1); /* BUG: No ordering vs. load from a!!! */
740 if (q) {
741 /* WRITE_ONCE(b, 1); -- moved up, BUG!!! */
742 do_something();
743 } else {
744 /* WRITE_ONCE(b, 1); -- moved up, BUG!!! */
745 do_something_else();
746 }
748 Now there is no conditional between the load from 'a' and the store to
749 'b', which means that the CPU is within its rights to reorder them:
750 The conditional is absolutely required, and must be present in the
751 assembly code even after all compiler optimizations have been applied.
752 Therefore, if you need ordering in this example, you need explicit
753 memory barriers, for example, smp_store_release():
755 q = READ_ONCE(a);
756 if (q) {
757 smp_store_release(&b, 1);
758 do_something();
759 } else {
760 smp_store_release(&b, 1);
761 do_something_else();
762 }
764 In contrast, without explicit memory barriers, two-legged-if control
765 ordering is guaranteed only when the stores differ, for example:
767 q = READ_ONCE(a);
768 if (q) {
769 WRITE_ONCE(b, 1);
770 do_something();
771 } else {
772 WRITE_ONCE(b, 2);
773 do_something_else();
774 }
776 The initial READ_ONCE() is still required to prevent the compiler from
777 proving the value of 'a'.
779 In addition, you need to be careful what you do with the local variable 'q',
780 otherwise the compiler might be able to guess the value and again remove
781 the needed conditional. For example:
783 q = READ_ONCE(a);
784 if (q % MAX) {
785 WRITE_ONCE(b, 1);
786 do_something();
787 } else {
788 WRITE_ONCE(b, 2);
789 do_something_else();
790 }
792 If MAX is defined to be 1, then the compiler knows that (q % MAX) is
793 equal to zero, in which case the compiler is within its rights to
794 transform the above code into the following:
796 q = READ_ONCE(a);
797 WRITE_ONCE(b, 2);
798 do_something_else();
800 Given this transformation, the CPU is not required to respect the ordering
801 between the load from variable 'a' and the store to variable 'b'. It is
802 tempting to add a barrier(), but this does not help. The conditional
803 is gone, and the barrier won't bring it back. Therefore, if you are
804 relying on this ordering, you should make sure that MAX is greater than
805 one, perhaps as follows:
807 q = READ_ONCE(a);
808 BUILD_BUG_ON(MAX <= 1); /* Order load from a with store to b. */
809 if (q % MAX) {
810 WRITE_ONCE(b, 1);
811 do_something();
812 } else {
813 WRITE_ONCE(b, 2);
814 do_something_else();
815 }
817 Please note once again that the stores to 'b' differ. If they were
818 identical, as noted earlier, the compiler could pull this store outside
819 of the 'if' statement.
821 You must also be careful not to rely too much on boolean short-circuit
822 evaluation. Consider this example:
824 q = READ_ONCE(a);
825 if (q || 1 > 0)
826 WRITE_ONCE(b, 1);
828 Because the first condition cannot fault and the second condition is
829 always true, the compiler can transform this example as following,
830 defeating control dependency:
832 q = READ_ONCE(a);
833 WRITE_ONCE(b, 1);
835 This example underscores the need to ensure that the compiler cannot
836 out-guess your code. More generally, although READ_ONCE() does force
837 the compiler to actually emit code for a given load, it does not force
838 the compiler to use the results.
840 In addition, control dependencies apply only to the then-clause and
841 else-clause of the if-statement in question. In particular, it does
842 not necessarily apply to code following the if-statement:
844 q = READ_ONCE(a);
845 if (q) {
846 WRITE_ONCE(b, 1);
847 } else {
848 WRITE_ONCE(b, 2);
849 }
850 WRITE_ONCE(c, 1); /* BUG: No ordering against the read from 'a'. */
852 It is tempting to argue that there in fact is ordering because the
853 compiler cannot reorder volatile accesses and also cannot reorder
854 the writes to 'b' with the condition. Unfortunately for this line
855 of reasoning, the compiler might compile the two writes to 'b' as
856 conditional-move instructions, as in this fanciful pseudo-assembly
857 language:
859 ld r1,a
860 cmp r1,$0
861 cmov,ne r4,$1
862 cmov,eq r4,$2
863 st r4,b
864 st $1,c
866 A weakly ordered CPU would have no dependency of any sort between the load
867 from 'a' and the store to 'c'. The control dependencies would extend
868 only to the pair of cmov instructions and the store depending on them.
869 In short, control dependencies apply only to the stores in the then-clause
870 and else-clause of the if-statement in question (including functions
871 invoked by those two clauses), not to code following that if-statement.
874 Note well that the ordering provided by a control dependency is local
875 to the CPU containing it. See the section on "Multicopy atomicity"
876 for more information.
879 In summary:
881 (*) Control dependencies can order prior loads against later stores.
882 However, they do -not- guarantee any other sort of ordering:
883 Not prior loads against later loads, nor prior stores against
884 later anything. If you need these other forms of ordering,
885 use smp_rmb(), smp_wmb(), or, in the case of prior stores and
886 later loads, smp_mb().
888 (*) If both legs of the "if" statement begin with identical stores to
889 the same variable, then those stores must be ordered, either by
890 preceding both of them with smp_mb() or by using smp_store_release()
891 to carry out the stores. Please note that it is -not- sufficient
892 to use barrier() at beginning of each leg of the "if" statement
893 because, as shown by the example above, optimizing compilers can
894 destroy the control dependency while respecting the letter of the
895 barrier() law.
897 (*) Control dependencies require at least one run-time conditional
898 between the prior load and the subsequent store, and this
899 conditional must involve the prior load. If the compiler is able
900 to optimize the conditional away, it will have also optimized
901 away the ordering. Careful use of READ_ONCE() and WRITE_ONCE()
902 can help to preserve the needed conditional.
904 (*) Control dependencies require that the compiler avoid reordering the
905 dependency into nonexistence. Careful use of READ_ONCE() or
906 atomic{,64}_read() can help to preserve your control dependency.
907 Please see the COMPILER BARRIER section for more information.
909 (*) Control dependencies apply only to the then-clause and else-clause
910 of the if-statement containing the control dependency, including
911 any functions that these two clauses call. Control dependencies
912 do -not- apply to code following the if-statement containing the
913 control dependency.
915 (*) Control dependencies pair normally with other types of barriers.
917 (*) Control dependencies do -not- provide multicopy atomicity. If you
918 need all the CPUs to see a given store at the same time, use smp_mb().
920 (*) Compilers do not understand control dependencies. It is therefore
921 your job to ensure that they do not break your code.
924 SMP BARRIER PAIRING
925 -------------------
927 When dealing with CPU-CPU interactions, certain types of memory barrier should
928 always be paired. A lack of appropriate pairing is almost certainly an error.
930 General barriers pair with each other, though they also pair with most
931 other types of barriers, albeit without multicopy atomicity. An acquire
932 barrier pairs with a release barrier, but both may also pair with other
933 barriers, including of course general barriers. A write barrier pairs
934 with a data dependency barrier, a control dependency, an acquire barrier,
935 a release barrier, a read barrier, or a general barrier. Similarly a
936 read barrier, control dependency, or a data dependency barrier pairs
937 with a write barrier, an acquire barrier, a release barrier, or a
938 general barrier:
940 CPU 1 CPU 2
941 =============== ===============
942 WRITE_ONCE(a, 1);
943 <write barrier>
944 WRITE_ONCE(b, 2); x = READ_ONCE(b);
945 <read barrier>
946 y = READ_ONCE(a);
948 Or:
950 CPU 1 CPU 2
951 =============== ===============================
952 a = 1;
953 <write barrier>
954 WRITE_ONCE(b, &a); x = READ_ONCE(b);
955 <data dependency barrier>
956 y = *x;
958 Or even:
960 CPU 1 CPU 2
961 =============== ===============================
962 r1 = READ_ONCE(y);
963 <general barrier>
964 WRITE_ONCE(x, 1); if (r2 = READ_ONCE(x)) {
965 <implicit control dependency>
966 WRITE_ONCE(y, 1);
967 }
969 assert(r1 == 0 || r2 == 0);
971 Basically, the read barrier always has to be there, even though it can be of
972 the "weaker" type.
974 [!] Note that the stores before the write barrier would normally be expected to
975 match the loads after the read barrier or the data dependency barrier, and vice
976 versa:
978 CPU 1 CPU 2
979 =================== ===================
980 WRITE_ONCE(a, 1); }---- --->{ v = READ_ONCE(c);
981 WRITE_ONCE(b, 2); } \ / { w = READ_ONCE(d);
982 <write barrier> \ <read barrier>
983 WRITE_ONCE(c, 3); } / \ { x = READ_ONCE(a);
984 WRITE_ONCE(d, 4); }---- --->{ y = READ_ONCE(b);
987 EXAMPLES OF MEMORY BARRIER SEQUENCES
988 ------------------------------------
990 Firstly, write barriers act as partial orderings on store operations.
991 Consider the following sequence of events:
993 CPU 1
994 =======================
995 STORE A = 1
996 STORE B = 2
997 STORE C = 3
998 <write barrier>
999 STORE D = 4
1000 STORE E = 5
1002 This sequence of events is committed to the memory coherence system in an order
1003 that the rest of the system might perceive as the unordered set of { STORE A,
1004 STORE B, STORE C } all occurring before the unordered set of { STORE D, STORE E
1005 }:
1007 +-------+ : :
1008 | | +------+
1009 | |------>| C=3 | } /\
1010 | | : +------+ }----- \ -----> Events perceptible to
1011 | | : | A=1 | } \/ the rest of the system
1012 | | : +------+ }
1013 | CPU 1 | : | B=2 | }
1014 | | +------+ }
1015 | | wwwwwwwwwwwwwwww } <--- At this point the write barrier
1016 | | +------+ } requires all stores prior to the
1017 | | : | E=5 | } barrier to be committed before
1018 | | : +------+ } further stores may take place
1019 | |------>| D=4 | }
1020 | | +------+
1021 +-------+ : :
1022 |
1023 | Sequence in which stores are committed to the
1024 | memory system by CPU 1
1025 V
1028 Secondly, data dependency barriers act as partial orderings on data-dependent
1029 loads. Consider the following sequence of events:
1031 CPU 1 CPU 2
1032 ======================= =======================
1033 { B = 7; X = 9; Y = 8; C = &Y }
1034 STORE A = 1
1035 STORE B = 2
1036 <write barrier>
1037 STORE C = &B LOAD X
1038 STORE D = 4 LOAD C (gets &B)
1039 LOAD *C (reads B)
1041 Without intervention, CPU 2 may perceive the events on CPU 1 in some
1042 effectively random order, despite the write barrier issued by CPU 1:
1044 +-------+ : : : :
1045 | | +------+ +-------+ | Sequence of update
1046 | |------>| B=2 |----- --->| Y->8 | | of perception on
1047 | | : +------+ \ +-------+ | CPU 2
1048 | CPU 1 | : | A=1 | \ --->| C->&Y | V
1049 | | +------+ | +-------+
1050 | | wwwwwwwwwwwwwwww | : :
1051 | | +------+ | : :
1052 | | : | C=&B |--- | : : +-------+
1053 | | : +------+ \ | +-------+ | |
1054 | |------>| D=4 | ----------->| C->&B |------>| |
1055 | | +------+ | +-------+ | |
1056 +-------+ : : | : : | |
1057 | : : | |
1058 | : : | CPU 2 |
1059 | +-------+ | |
1060 Apparently incorrect ---> | | B->7 |------>| |
1061 perception of B (!) | +-------+ | |
1062 | : : | |
1063 | +-------+ | |
1064 The load of X holds ---> \ | X->9 |------>| |
1065 up the maintenance \ +-------+ | |
1066 of coherence of B ----->| B->2 | +-------+
1067 +-------+
1068 : :
1071 In the above example, CPU 2 perceives that B is 7, despite the load of *C
1072 (which would be B) coming after the LOAD of C.
1074 If, however, a data dependency barrier were to be placed between the load of C
1075 and the load of *C (ie: B) on CPU 2:
1077 CPU 1 CPU 2
1078 ======================= =======================
1079 { B = 7; X = 9; Y = 8; C = &Y }
1080 STORE A = 1
1081 STORE B = 2
1082 <write barrier>
1083 STORE C = &B LOAD X
1084 STORE D = 4 LOAD C (gets &B)
1085 <data dependency barrier>
1086 LOAD *C (reads B)
1088 then the following will occur:
1090 +-------+ : : : :
1091 | | +------+ +-------+
1092 | |------>| B=2 |----- --->| Y->8 |
1093 | | : +------+ \ +-------+
1094 | CPU 1 | : | A=1 | \ --->| C->&Y |
1095 | | +------+ | +-------+
1096 | | wwwwwwwwwwwwwwww | : :
1097 | | +------+ | : :
1098 | | : | C=&B |--- | : : +-------+
1099 | | : +------+ \ | +-------+ | |
1100 | |------>| D=4 | ----------->| C->&B |------>| |
1101 | | +------+ | +-------+ | |
1102 +-------+ : : | : : | |
1103 | : : | |
1104 | : : | CPU 2 |
1105 | +-------+ | |
1106 | | X->9 |------>| |
1107 | +-------+ | |
1108 Makes sure all effects ---> \ ddddddddddddddddd | |
1109 prior to the store of C \ +-------+ | |
1110 are perceptible to ----->| B->2 |------>| |
1111 subsequent loads +-------+ | |
1112 : : +-------+
1115 And thirdly, a read barrier acts as a partial order on loads. Consider the
1116 following sequence of events:
1118 CPU 1 CPU 2
1119 ======================= =======================
1120 { A = 0, B = 9 }
1121 STORE A=1
1122 <write barrier>
1123 STORE B=2
1124 LOAD B
1125 LOAD A
1127 Without intervention, CPU 2 may then choose to perceive the events on CPU 1 in
1128 some effectively random order, despite the write barrier issued by CPU 1:
1130 +-------+ : : : :
1131 | | +------+ +-------+
1132 | |------>| A=1 |------ --->| A->0 |
1133 | | +------+ \ +-------+
1134 | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 |
1135 | | +------+ | +-------+
1136 | |------>| B=2 |--- | : :
1137 | | +------+ \ | : : +-------+
1138 +-------+ : : \ | +-------+ | |
1139 ---------->| B->2 |------>| |
1140 | +-------+ | CPU 2 |
1141 | | A->0 |------>| |
1142 | +-------+ | |
1143 | : : +-------+
1144 \ : :
1145 \ +-------+
1146 ---->| A->1 |
1147 +-------+
1148 : :
1151 If, however, a read barrier were to be placed between the load of B and the
1152 load of A on CPU 2:
1154 CPU 1 CPU 2
1155 ======================= =======================
1156 { A = 0, B = 9 }
1157 STORE A=1
1158 <write barrier>
1159 STORE B=2
1160 LOAD B
1161 <read barrier>
1162 LOAD A
1164 then the partial ordering imposed by CPU 1 will be perceived correctly by CPU
1165 2:
1167 +-------+ : : : :
1168 | | +------+ +-------+
1169 | |------>| A=1 |------ --->| A->0 |
1170 | | +------+ \ +-------+
1171 | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 |
1172 | | +------+ | +-------+
1173 | |------>| B=2 |--- | : :
1174 | | +------+ \ | : : +-------+
1175 +-------+ : : \ | +-------+ | |
1176 ---------->| B->2 |------>| |
1177 | +-------+ | CPU 2 |
1178 | : : | |
1179 | : : | |
1180 At this point the read ----> \ rrrrrrrrrrrrrrrrr | |
1181 barrier causes all effects \ +-------+ | |
1182 prior to the storage of B ---->| A->1 |------>| |
1183 to be perceptible to CPU 2 +-------+ | |
1184 : : +-------+
1187 To illustrate this more completely, consider what could happen if the code
1188 contained a load of A either side of the read barrier:
1190 CPU 1 CPU 2
1191 ======================= =======================
1192 { A = 0, B = 9 }
1193 STORE A=1
1194 <write barrier>
1195 STORE B=2
1196 LOAD B
1197 LOAD A [first load of A]
1198 <read barrier>
1199 LOAD A [second load of A]
1201 Even though the two loads of A both occur after the load of B, they may both
1202 come up with different values:
1204 +-------+ : : : :
1205 | | +------+ +-------+
1206 | |------>| A=1 |------ --->| A->0 |
1207 | | +------+ \ +-------+
1208 | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 |
1209 | | +------+ | +-------+
1210 | |------>| B=2 |--- | : :
1211 | | +------+ \ | : : +-------+
1212 +-------+ : : \ | +-------+ | |
1213 ---------->| B->2 |------>| |
1214 | +-------+ | CPU 2 |
1215 | : : | |
1216 | : : | |
1217 | +-------+ | |
1218 | | A->0 |------>| 1st |
1219 | +-------+ | |
1220 At this point the read ----> \ rrrrrrrrrrrrrrrrr | |
1221 barrier causes all effects \ +-------+ | |
1222 prior to the storage of B ---->| A->1 |------>| 2nd |
1223 to be perceptible to CPU 2 +-------+ | |
1224 : : +-------+
1227 But it may be that the update to A from CPU 1 becomes perceptible to CPU 2
1228 before the read barrier completes anyway:
1230 +-------+ : : : :
1231 | | +------+ +-------+
1232 | |------>| A=1 |------ --->| A->0 |
1233 | | +------+ \ +-------+
1234 | CPU 1 | wwwwwwwwwwwwwwww \ --->| B->9 |
1235 | | +------+ | +-------+
1236 | |------>| B=2 |--- | : :
1237 | | +------+ \ | : : +-------+
1238 +-------+ : : \ | +-------+ | |
1239 ---------->| B->2 |------>| |
1240 | +-------+ | CPU 2 |
1241 | : : | |
1242 \ : : | |
1243 \ +-------+ | |
1244 ---->| A->1 |------>| 1st |
1245 +-------+ | |
1246 rrrrrrrrrrrrrrrrr | |
1247 +-------+ | |
1248 | A->1 |------>| 2nd |
1249 +-------+ | |
1250 : : +-------+
1253 The guarantee is that the second load will always come up with A == 1 if the
1254 load of B came up with B == 2. No such guarantee exists for the first load of
1255 A; that may come up with either A == 0 or A == 1.
1258 READ MEMORY BARRIERS VS LOAD SPECULATION
1259 ----------------------------------------
1261 Many CPUs speculate with loads: that is they see that they will need to load an
1262 item from memory, and they find a time where they're not using the bus for any
1263 other loads, and so do the load in advance - even though they haven't actually
1264 got to that point in the instruction execution flow yet. This permits the
1265 actual load instruction to potentially complete immediately because the CPU
1266 already has the value to hand.
1268 It may turn out that the CPU didn't actually need the value - perhaps because a
1269 branch circumvented the load - in which case it can discard the value or just
1270 cache it for later use.
1272 Consider:
1274 CPU 1 CPU 2
1275 ======================= =======================
1276 LOAD B
1277 DIVIDE } Divide instructions generally
1278 DIVIDE } take a long time to perform
1279 LOAD A
1281 Which might appear as this:
1283 : : +-------+
1284 +-------+ | |
1285 --->| B->2 |------>| |
1286 +-------+ | CPU 2 |
1287 : :DIVIDE | |
1288 +-------+ | |
1289 The CPU being busy doing a ---> --->| A->0 |~~~~ | |
1290 division speculates on the +-------+ ~ | |
1291 LOAD of A : : ~ | |
1292 : :DIVIDE | |
1293 : : ~ | |
1294 Once the divisions are complete --> : : ~-->| |
1295 the CPU can then perform the : : | |
1296 LOAD with immediate effect : : +-------+
1299 Placing a read barrier or a data dependency barrier just before the second
1300 load:
1302 CPU 1 CPU 2
1303 ======================= =======================
1304 LOAD B
1305 DIVIDE
1306 DIVIDE
1307 <read barrier>
1308 LOAD A
1310 will force any value speculatively obtained to be reconsidered to an extent
1311 dependent on the type of barrier used. If there was no change made to the
1312 speculated memory location, then the speculated value will just be used:
1314 : : +-------+
1315 +-------+ | |
1316 --->| B->2 |------>| |
1317 +-------+ | CPU 2 |
1318 : :DIVIDE | |
1319 +-------+ | |
1320 The CPU being busy doing a ---> --->| A->0 |~~~~ | |
1321 division speculates on the +-------+ ~ | |
1322 LOAD of A : : ~ | |
1323 : :DIVIDE | |
1324 : : ~ | |
1325 : : ~ | |
1326 rrrrrrrrrrrrrrrr~ | |
1327 : : ~ | |
1328 : : ~-->| |
1329 : : | |
1330 : : +-------+
1333 but if there was an update or an invalidation from another CPU pending, then
1334 the speculation will be cancelled and the value reloaded:
1336 : : +-------+
1337 +-------+ | |
1338 --->| B->2 |------>| |
1339 +-------+ | CPU 2 |
1340 : :DIVIDE | |
1341 +-------+ | |
1342 The CPU being busy doing a ---> --->| A->0 |~~~~ | |
1343 division speculates on the +-------+ ~ | |
1344 LOAD of A : : ~ | |
1345 : :DIVIDE | |
1346 : : ~ | |
1347 : : ~ | |
1348 rrrrrrrrrrrrrrrrr | |
1349 +-------+ | |
1350 The speculation is discarded ---> --->| A->1 |------>| |
1351 and an updated value is +-------+ | |
1352 retrieved : : +-------+
1355 MULTICOPY ATOMICITY
1356 --------------------
1358 Multicopy atomicity is a deeply intuitive notion about ordering that is
1359 not always provided by real computer systems, namely that a given store
1360 becomes visible at the same time to all CPUs, or, alternatively, that all
1361 CPUs agree on the order in which all stores become visible. However,
1362 support of full multicopy atomicity would rule out valuable hardware
1363 optimizations, so a weaker form called ``other multicopy atomicity''
1364 instead guarantees only that a given store becomes visible at the same
1365 time to all -other- CPUs. The remainder of this document discusses this
1366 weaker form, but for brevity will call it simply ``multicopy atomicity''.
1368 The following example demonstrates multicopy atomicity:
1370 CPU 1 CPU 2 CPU 3
1371 ======================= ======================= =======================
1372 { X = 0, Y = 0 }
1373 STORE X=1 r1=LOAD X (reads 1) LOAD Y (reads 1)
1374 <general barrier> <read barrier>
1375 STORE Y=r1 LOAD X
1377 Suppose that CPU 2's load from X returns 1, which it then stores to Y,
1378 and CPU 3's load from Y returns 1. This indicates that CPU 1's store
1379 to X precedes CPU 2's load from X and that CPU 2's store to Y precedes
1380 CPU 3's load from Y. In addition, the memory barriers guarantee that
1381 CPU 2 executes its load before its store, and CPU 3 loads from Y before
1382 it loads from X. The question is then "Can CPU 3's load from X return 0?"
1384 Because CPU 3's load from X in some sense comes after CPU 2's load, it
1385 is natural to expect that CPU 3's load from X must therefore return 1.
1386 This expectation follows from multicopy atomicity: if a load executing
1387 on CPU B follows a load from the same variable executing on CPU A (and
1388 CPU A did not originally store the value which it read), then on
1389 multicopy-atomic systems, CPU B's load must return either the same value
1390 that CPU A's load did or some later value. However, the Linux kernel
1391 does not require systems to be multicopy atomic.
1393 The use of a general memory barrier in the example above compensates
1394 for any lack of multicopy atomicity. In the example, if CPU 2's load
1395 from X returns 1 and CPU 3's load from Y returns 1, then CPU 3's load
1396 from X must indeed also return 1.
1398 However, dependencies, read barriers, and write barriers are not always
1399 able to compensate for non-multicopy atomicity. For example, suppose
1400 that CPU 2's general barrier is removed from the above example, leaving
1401 only the data dependency shown below:
1403 CPU 1 CPU 2 CPU 3
1404 ======================= ======================= =======================
1405 { X = 0, Y = 0 }
1406 STORE X=1 r1=LOAD X (reads 1) LOAD Y (reads 1)
1407 <data dependency> <read barrier>
1408 STORE Y=r1 LOAD X (reads 0)
1410 This substitution allows non-multicopy atomicity to run rampant: in
1411 this example, it is perfectly legal for CPU 2's load from X to return 1,
1412 CPU 3's load from Y to return 1, and its load from X to return 0.
1414 The key point is that although CPU 2's data dependency orders its load
1415 and store, it does not guarantee to order CPU 1's store. Thus, if this
1416 example runs on a non-multicopy-atomic system where CPUs 1 and 2 share a
1417 store buffer or a level of cache, CPU 2 might have early access to CPU 1's
1418 writes. General barriers are therefore required to ensure that all CPUs
1419 agree on the combined order of multiple accesses.
1421 General barriers can compensate not only for non-multicopy atomicity,
1422 but can also generate additional ordering that can ensure that -all-
1423 CPUs will perceive the same order of -all- operations. In contrast, a
1424 chain of release-acquire pairs do not provide this additional ordering,
1425 which means that only those CPUs on the chain are guaranteed to agree
1426 on the combined order of the accesses. For example, switching to C code
1427 in deference to the ghost of Herman Hollerith:
1429 int u, v, x, y, z;
1431 void cpu0(void)
1432 {
1433 r0 = smp_load_acquire(&x);
1434 WRITE_ONCE(u, 1);
1435 smp_store_release(&y, 1);
1436 }
1438 void cpu1(void)
1439 {
1440 r1 = smp_load_acquire(&y);
1441 r4 = READ_ONCE(v);
1442 r5 = READ_ONCE(u);
1443 smp_store_release(&z, 1);
1444 }
1446 void cpu2(void)
1447 {
1448 r2 = smp_load_acquire(&z);
1449 smp_store_release(&x, 1);
1450 }
1452 void cpu3(void)
1453 {
1454 WRITE_ONCE(v, 1);
1455 smp_mb();
1456 r3 = READ_ONCE(u);
1457 }
1459 Because cpu0(), cpu1(), and cpu2() participate in a chain of
1460 smp_store_release()/smp_load_acquire() pairs, the following outcome
1461 is prohibited:
1463 r0 == 1 && r1 == 1 && r2 == 1
1465 Furthermore, because of the release-acquire relationship between cpu0()
1466 and cpu1(), cpu1() must see cpu0()'s writes, so that the following
1467 outcome is prohibited:
1469 r1 == 1 && r5 == 0
1471 However, the ordering provided by a release-acquire chain is local
1472 to the CPUs participating in that chain and does not apply to cpu3(),
1473 at least aside from stores. Therefore, the following outcome is possible:
1475 r0 == 0 && r1 == 1 && r2 == 1 && r3 == 0 && r4 == 0
1477 As an aside, the following outcome is also possible:
1479 r0 == 0 && r1 == 1 && r2 == 1 && r3 == 0 && r4 == 0 && r5 == 1
1481 Although cpu0(), cpu1(), and cpu2() will see their respective reads and
1482 writes in order, CPUs not involved in the release-acquire chain might
1483 well disagree on the order. This disagreement stems from the fact that
1484 the weak memory-barrier instructions used to implement smp_load_acquire()
1485 and smp_store_release() are not required to order prior stores against
1486 subsequent loads in all cases. This means that cpu3() can see cpu0()'s
1487 store to u as happening -after- cpu1()'s load from v, even though
1488 both cpu0() and cpu1() agree that these two operations occurred in the
1489 intended order.
1491 However, please keep in mind that smp_load_acquire() is not magic.
1492 In particular, it simply reads from its argument with ordering. It does
1493 -not- ensure that any particular value will be read. Therefore, the
1494 following outcome is possible:
1496 r0 == 0 && r1 == 0 && r2 == 0 && r5 == 0
1498 Note that this outcome can happen even on a mythical sequentially
1499 consistent system where nothing is ever reordered.
1501 To reiterate, if your code requires full ordering of all operations,
1502 use general barriers throughout.
1505 ========================
1506 EXPLICIT KERNEL BARRIERS
1507 ========================
1509 The Linux kernel has a variety of different barriers that act at different
1510 levels:
1512 (*) Compiler barrier.
1514 (*) CPU memory barriers.
1516 (*) MMIO write barrier.
1519 COMPILER BARRIER
1520 ----------------
1522 The Linux kernel has an explicit compiler barrier function that prevents the
1523 compiler from moving the memory accesses either side of it to the other side:
1525 barrier();
1527 This is a general barrier -- there are no read-read or write-write
1528 variants of barrier(). However, READ_ONCE() and WRITE_ONCE() can be
1529 thought of as weak forms of barrier() that affect only the specific
1530 accesses flagged by the READ_ONCE() or WRITE_ONCE().
1532 The barrier() function has the following effects:
1534 (*) Prevents the compiler from reordering accesses following the
1535 barrier() to precede any accesses preceding the barrier().
1536 One example use for this property is to ease communication between
1537 interrupt-handler code and the code that was interrupted.
1539 (*) Within a loop, forces the compiler to load the variables used
1540 in that loop's conditional on each pass through that loop.
1542 The READ_ONCE() and WRITE_ONCE() functions can prevent any number of
1543 optimizations that, while perfectly safe in single-threaded code, can
1544 be fatal in concurrent code. Here are some examples of these sorts
1545 of optimizations:
1547 (*) The compiler is within its rights to reorder loads and stores
1548 to the same variable, and in some cases, the CPU is within its
1549 rights to reorder loads to the same variable. This means that
1550 the following code:
1552 a[0] = x;
1553 a[1] = x;
1555 Might result in an older value of x stored in a[1] than in a[0].
1556 Prevent both the compiler and the CPU from doing this as follows:
1558 a[0] = READ_ONCE(x);
1559 a[1] = READ_ONCE(x);
1561 In short, READ_ONCE() and WRITE_ONCE() provide cache coherence for
1562 accesses from multiple CPUs to a single variable.
1564 (*) The compiler is within its rights to merge successive loads from
1565 the same variable. Such merging can cause the compiler to "optimize"
1566 the following code:
1568 while (tmp = a)
1569 do_something_with(tmp);
1571 into the following code, which, although in some sense legitimate
1572 for single-threaded code, is almost certainly not what the developer
1573 intended:
1575 if (tmp = a)
1576 for (;;)
1577 do_something_with(tmp);
1579 Use READ_ONCE() to prevent the compiler from doing this to you:
1581 while (tmp = READ_ONCE(a))
1582 do_something_with(tmp);
1584 (*) The compiler is within its rights to reload a variable, for example,
1585 in cases where high register pressure prevents the compiler from
1586 keeping all data of interest in registers. The compiler might
1587 therefore optimize the variable 'tmp' out of our previous example:
1589 while (tmp = a)
1590 do_something_with(tmp);
1592 This could result in the following code, which is perfectly safe in
1593 single-threaded code, but can be fatal in concurrent code:
1595 while (a)
1596 do_something_with(a);
1598 For example, the optimized version of this code could result in
1599 passing a zero to do_something_with() in the case where the variable
1600 a was modified by some other CPU between the "while" statement and
1601 the call to do_something_with().
1603 Again, use READ_ONCE() to prevent the compiler from doing this:
1605 while (tmp = READ_ONCE(a))
1606 do_something_with(tmp);
1608 Note that if the compiler runs short of registers, it might save
1609 tmp onto the stack. The overhead of this saving and later restoring
1610 is why compilers reload variables. Doing so is perfectly safe for
1611 single-threaded code, so you need to tell the compiler about cases
1612 where it is not safe.
1614 (*) The compiler is within its rights to omit a load entirely if it knows
1615 what the value will be. For example, if the compiler can prove that
1616 the value of variable 'a' is always zero, it can optimize this code:
1618 while (tmp = a)
1619 do_something_with(tmp);
1621 Into this:
1623 do { } while (0);
1625 This transformation is a win for single-threaded code because it
1626 gets rid of a load and a branch. The problem is that the compiler
1627 will carry out its proof assuming that the current CPU is the only
1628 one updating variable 'a'. If variable 'a' is shared, then the
1629 compiler's proof will be erroneous. Use READ_ONCE() to tell the
1630 compiler that it doesn't know as much as it thinks it does:
1632 while (tmp = READ_ONCE(a))
1633 do_something_with(tmp);
1635 But please note that the compiler is also closely watching what you
1636 do with the value after the READ_ONCE(). For example, suppose you
1637 do the following and MAX is a preprocessor macro with the value 1:
1639 while ((tmp = READ_ONCE(a)) % MAX)
1640 do_something_with(tmp);
1642 Then the compiler knows that the result of the "%" operator applied
1643 to MAX will always be zero, again allowing the compiler to optimize
1644 the code into near-nonexistence. (It will still load from the
1645 variable 'a'.)
1647 (*) Similarly, the compiler is within its rights to omit a store entirely
1648 if it knows that the variable already has the value being stored.
1649 Again, the compiler assumes that the current CPU is the only one
1650 storing into the variable, which can cause the compiler to do the
1651 wrong thing for shared variables. For example, suppose you have
1652 the following:
1654 a = 0;
1655 ... Code that does not store to variable a ...
1656 a = 0;
1658 The compiler sees that the value of variable 'a' is already zero, so
1659 it might well omit the second store. This would come as a fatal
1660 surprise if some other CPU might have stored to variable 'a' in the
1661 meantime.
1663 Use WRITE_ONCE() to prevent the compiler from making this sort of
1664 wrong guess:
1666 WRITE_ONCE(a, 0);
1667 ... Code that does not store to variable a ...
1668 WRITE_ONCE(a, 0);
1670 (*) The compiler is within its rights to reorder memory accesses unless
1671 you tell it not to. For example, consider the following interaction
1672 between process-level code and an interrupt handler:
1674 void process_level(void)
1675 {
1676 msg = get_message();
1677 flag = true;
1678 }
1680 void interrupt_handler(void)
1681 {
1682 if (flag)
1683 process_message(msg);
1684 }
1686 There is nothing to prevent the compiler from transforming
1687 process_level() to the following, in fact, this might well be a
1688 win for single-threaded code:
1690 void process_level(void)
1691 {
1692 flag = true;
1693 msg = get_message();
1694 }
1696 If the interrupt occurs between these two statement, then
1697 interrupt_handler() might be passed a garbled msg. Use WRITE_ONCE()
1698 to prevent this as follows:
1700 void process_level(void)
1701 {
1702 WRITE_ONCE(msg, get_message());
1703 WRITE_ONCE(flag, true);
1704 }
1706 void interrupt_handler(void)
1707 {
1708 if (READ_ONCE(flag))
1709 process_message(READ_ONCE(msg));
1710 }
1712 Note that the READ_ONCE() and WRITE_ONCE() wrappers in
1713 interrupt_handler() are needed if this interrupt handler can itself
1714 be interrupted by something that also accesses 'flag' and 'msg',
1715 for example, a nested interrupt or an NMI. Otherwise, READ_ONCE()
1716 and WRITE_ONCE() are not needed in interrupt_handler() other than
1717 for documentation purposes. (Note also that nested interrupts
1718 do not typically occur in modern Linux kernels, in fact, if an
1719 interrupt handler returns with interrupts enabled, you will get a
1720 WARN_ONCE() splat.)
1722 You should assume that the compiler can move READ_ONCE() and
1723 WRITE_ONCE() past code not containing READ_ONCE(), WRITE_ONCE(),
1724 barrier(), or similar primitives.
1726 This effect could also be achieved using barrier(), but READ_ONCE()
1727 and WRITE_ONCE() are more selective: With READ_ONCE() and
1728 WRITE_ONCE(), the compiler need only forget the contents of the
1729 indicated memory locations, while with barrier() the compiler must
1730 discard the value of all memory locations that it has currented
1731 cached in any machine registers. Of course, the compiler must also
1732 respect the order in which the READ_ONCE()s and WRITE_ONCE()s occur,
1733 though the CPU of course need not do so.
1735 (*) The compiler is within its rights to invent stores to a variable,
1736 as in the following example:
1738 if (a)
1739 b = a;
1740 else
1741 b = 42;
1743 The compiler might save a branch by optimizing this as follows:
1745 b = 42;
1746 if (a)
1747 b = a;
1749 In single-threaded code, this is not only safe, but also saves
1750 a branch. Unfortunately, in concurrent code, this optimization
1751 could cause some other CPU to see a spurious value of 42 -- even
1752 if variable 'a' was never zero -- when loading variable 'b'.
1753 Use WRITE_ONCE() to prevent this as follows:
1755 if (a)
1756 WRITE_ONCE(b, a);
1757 else
1758 WRITE_ONCE(b, 42);
1760 The compiler can also invent loads. These are usually less
1761 damaging, but they can result in cache-line bouncing and thus in
1762 poor performance and scalability. Use READ_ONCE() to prevent
1763 invented loads.
1765 (*) For aligned memory locations whose size allows them to be accessed
1766 with a single memory-reference instruction, prevents "load tearing"
1767 and "store tearing," in which a single large access is replaced by
1768 multiple smaller accesses. For example, given an architecture having
1769 16-bit store instructions with 7-bit immediate fields, the compiler
1770 might be tempted to use two 16-bit store-immediate instructions to
1771 implement the following 32-bit store:
1773 p = 0x00010002;
1775 Please note that GCC really does use this sort of optimization,
1776 which is not surprising given that it would likely take more
1777 than two instructions to build the constant and then store it.
1778 This optimization can therefore be a win in single-threaded code.
1779 In fact, a recent bug (since fixed) caused GCC to incorrectly use
1780 this optimization in a volatile store. In the absence of such bugs,
1781 use of WRITE_ONCE() prevents store tearing in the following example:
1783 WRITE_ONCE(p, 0x00010002);
1785 Use of packed structures can also result in load and store tearing,
1786 as in this example:
1788 struct __attribute__((__packed__)) foo {
1789 short a;
1790 int b;
1791 short c;
1792 };
1793 struct foo foo1, foo2;
1794 ...
1796 foo2.a = foo1.a;
1797 foo2.b = foo1.b;
1798 foo2.c = foo1.c;
1800 Because there are no READ_ONCE() or WRITE_ONCE() wrappers and no
1801 volatile markings, the compiler would be well within its rights to
1802 implement these three assignment statements as a pair of 32-bit
1803 loads followed by a pair of 32-bit stores. This would result in
1804 load tearing on 'foo1.b' and store tearing on 'foo2.b'. READ_ONCE()
1805 and WRITE_ONCE() again prevent tearing in this example:
1807 foo2.a = foo1.a;
1808 WRITE_ONCE(foo2.b, READ_ONCE(foo1.b));
1809 foo2.c = foo1.c;
1811 All that aside, it is never necessary to use READ_ONCE() and
1812 WRITE_ONCE() on a variable that has been marked volatile. For example,
1813 because 'jiffies' is marked volatile, it is never necessary to
1814 say READ_ONCE(jiffies). The reason for this is that READ_ONCE() and
1815 WRITE_ONCE() are implemented as volatile casts, which has no effect when
1816 its argument is already marked volatile.
1818 Please note that these compiler barriers have no direct effect on the CPU,
1819 which may then reorder things however it wishes.
1822 CPU MEMORY BARRIERS
1823 -------------------
1825 The Linux kernel has eight basic CPU memory barriers:
1827 TYPE MANDATORY SMP CONDITIONAL
1828 =============== ======================= ===========================
1829 GENERAL mb() smp_mb()
1830 WRITE wmb() smp_wmb()
1831 READ rmb() smp_rmb()
1832 DATA DEPENDENCY READ_ONCE()
1835 All memory barriers except the data dependency barriers imply a compiler
1836 barrier. Data dependencies do not impose any additional compiler ordering.
1838 Aside: In the case of data dependencies, the compiler would be expected
1839 to issue the loads in the correct order (eg. `a[b]` would have to load
1840 the value of b before loading a[b]), however there is no guarantee in
1841 the C specification that the compiler may not speculate the value of b
1842 (eg. is equal to 1) and load a before b (eg. tmp = a[1]; if (b != 1)
1843 tmp = a[b]; ). There is also the problem of a compiler reloading b after
1844 having loaded a[b], thus having a newer copy of b than a[b]. A consensus
1845 has not yet been reached about these problems, however the READ_ONCE()
1846 macro is a good place to start looking.
1848 SMP memory barriers are reduced to compiler barriers on uniprocessor compiled
1849 systems because it is assumed that a CPU will appear to be self-consistent,
1850 and will order overlapping accesses correctly with respect to itself.
1851 However, see the subsection on "Virtual Machine Guests" below.
1853 [!] Note that SMP memory barriers _must_ be used to control the ordering of
1854 references to shared memory on SMP systems, though the use of locking instead
1855 is sufficient.
1857 Mandatory barriers should not be used to control SMP effects, since mandatory
1858 barriers impose unnecessary overhead on both SMP and UP systems. They may,
1859 however, be used to control MMIO effects on accesses through relaxed memory I/O
1860 windows. These barriers are required even on non-SMP systems as they affect
1861 the order in which memory operations appear to a device by prohibiting both the
1862 compiler and the CPU from reordering them.
1865 There are some more advanced barrier functions:
1867 (*) smp_store_mb(var, value)
1869 This assigns the value to the variable and then inserts a full memory
1870 barrier after it. It isn't guaranteed to insert anything more than a
1871 compiler barrier in a UP compilation.
1874 (*) smp_mb__before_atomic();
1875 (*) smp_mb__after_atomic();
1877 These are for use with atomic (such as add, subtract, increment and
1878 decrement) functions that don't return a value, especially when used for
1879 reference counting. These functions do not imply memory barriers.
1881 These are also used for atomic bitop functions that do not return a
1882 value (such as set_bit and clear_bit).
1884 As an example, consider a piece of code that marks an object as being dead
1885 and then decrements the object's reference count:
1887 obj->dead = 1;
1888 smp_mb__before_atomic();
1889 atomic_dec(&obj->ref_count);
1891 This makes sure that the death mark on the object is perceived to be set
1892 *before* the reference counter is decremented.
1894 See Documentation/atomic_{t,bitops}.txt for more information.
1897 (*) dma_wmb();
1898 (*) dma_rmb();
1900 These are for use with consistent memory to guarantee the ordering
1901 of writes or reads of shared memory accessible to both the CPU and a
1902 DMA capable device.
1904 For example, consider a device driver that shares memory with a device
1905 and uses a descriptor status value to indicate if the descriptor belongs
1906 to the device or the CPU, and a doorbell to notify it when new
1907 descriptors are available:
1909 if (desc->status != DEVICE_OWN) {
1910 /* do not read data until we own descriptor */
1911 dma_rmb();
1913 /* read/modify data */
1914 read_data = desc->data;
1915 desc->data = write_data;
1917 /* flush modifications before status update */
1918 dma_wmb();
1920 /* assign ownership */
1921 desc->status = DEVICE_OWN;
1923 /* notify device of new descriptors */
1924 writel(DESC_NOTIFY, doorbell);
1925 }
1927 The dma_rmb() allows us guarantee the device has released ownership
1928 before we read the data from the descriptor, and the dma_wmb() allows
1929 us to guarantee the data is written to the descriptor before the device
1930 can see it now has ownership. Note that, when using writel(), a prior
1931 wmb() is not needed to guarantee that the cache coherent memory writes
1932 have completed before writing to the MMIO region. The cheaper
1933 writel_relaxed() does not provide this guarantee and must not be used
1934 here.
1936 See the subsection "Kernel I/O barrier effects" for more information on
1937 relaxed I/O accessors and the Documentation/DMA-API.txt file for more
1938 information on consistent memory.
1941 MMIO WRITE BARRIER
1942 ------------------
1944 The Linux kernel also has a special barrier for use with memory-mapped I/O
1945 writes:
1947 mmiowb();
1949 This is a variation on the mandatory write barrier that causes writes to weakly
1950 ordered I/O regions to be partially ordered. Its effects may go beyond the
1951 CPU->Hardware interface and actually affect the hardware at some level.
1953 See the subsection "Acquires vs I/O accesses" for more information.
1956 ===============================
1957 IMPLICIT KERNEL MEMORY BARRIERS
1958 ===============================
1960 Some of the other functions in the linux kernel imply memory barriers, amongst
1961 which are locking and scheduling functions.
1963 This specification is a _minimum_ guarantee; any particular architecture may
1964 provide more substantial guarantees, but these may not be relied upon outside
1965 of arch specific code.
1968 LOCK ACQUISITION FUNCTIONS
1969 --------------------------
1971 The Linux kernel has a number of locking constructs:
1973 (*) spin locks
1974 (*) R/W spin locks
1975 (*) mutexes
1976 (*) semaphores
1977 (*) R/W semaphores
1979 In all cases there are variants on "ACQUIRE" operations and "RELEASE" operations
1980 for each construct. These operations all imply certain barriers:
1982 (1) ACQUIRE operation implication:
1984 Memory operations issued after the ACQUIRE will be completed after the
1985 ACQUIRE operation has completed.
1987 Memory operations issued before the ACQUIRE may be completed after
1988 the ACQUIRE operation has completed.
1990 (2) RELEASE operation implication:
1992 Memory operations issued before the RELEASE will be completed before the
1993 RELEASE operation has completed.
1995 Memory operations issued after the RELEASE may be completed before the
1996 RELEASE operation has completed.
1998 (3) ACQUIRE vs ACQUIRE implication:
2000 All ACQUIRE operations issued before another ACQUIRE operation will be
2001 completed before that ACQUIRE operation.
2003 (4) ACQUIRE vs RELEASE implication:
2005 All ACQUIRE operations issued before a RELEASE operation will be
2006 completed before the RELEASE operation.
2008 (5) Failed conditional ACQUIRE implication:
2010 Certain locking variants of the ACQUIRE operation may fail, either due to
2011 being unable to get the lock immediately, or due to receiving an unblocked
2012 signal whilst asleep waiting for the lock to become available. Failed
2013 locks do not imply any sort of barrier.
2015 [!] Note: one of the consequences of lock ACQUIREs and RELEASEs being only
2016 one-way barriers is that the effects of instructions outside of a critical
2017 section may seep into the inside of the critical section.
2019 An ACQUIRE followed by a RELEASE may not be assumed to be full memory barrier
2020 because it is possible for an access preceding the ACQUIRE to happen after the
2021 ACQUIRE, and an access following the RELEASE to happen before the RELEASE, and
2022 the two accesses can themselves then cross:
2024 *A = a;
2025 ACQUIRE M
2026 RELEASE M
2027 *B = b;
2029 may occur as:
2031 ACQUIRE M, STORE *B, STORE *A, RELEASE M
2033 When the ACQUIRE and RELEASE are a lock acquisition and release,
2034 respectively, this same reordering can occur if the lock's ACQUIRE and
2035 RELEASE are to the same lock variable, but only from the perspective of
2036 another CPU not holding that lock. In short, a ACQUIRE followed by an
2037 RELEASE may -not- be assumed to be a full memory barrier.
2039 Similarly, the reverse case of a RELEASE followed by an ACQUIRE does
2040 not imply a full memory barrier. Therefore, the CPU's execution of the
2041 critical sections corresponding to the RELEASE and the ACQUIRE can cross,
2042 so that:
2044 *A = a;
2045 RELEASE M
2046 ACQUIRE N
2047 *B = b;
2049 could occur as:
2051 ACQUIRE N, STORE *B, STORE *A, RELEASE M
2053 It might appear that this reordering could introduce a deadlock.
2054 However, this cannot happen because if such a deadlock threatened,
2055 the RELEASE would simply complete, thereby avoiding the deadlock.
2057 Why does this work?
2059 One key point is that we are only talking about the CPU doing
2060 the reordering, not the compiler. If the compiler (or, for
2061 that matter, the developer) switched the operations, deadlock
2062 -could- occur.
2064 But suppose the CPU reordered the operations. In this case,
2065 the unlock precedes the lock in the assembly code. The CPU
2066 simply elected to try executing the later lock operation first.
2067 If there is a deadlock, this lock operation will simply spin (or
2068 try to sleep, but more on that later). The CPU will eventually
2069 execute the unlock operation (which preceded the lock operation
2070 in the assembly code), which will unravel the potential deadlock,
2071 allowing the lock operation to succeed.
2073 But what if the lock is a sleeplock? In that case, the code will
2074 try to enter the scheduler, where it will eventually encounter
2075 a memory barrier, which will force the earlier unlock operation
2076 to complete, again unraveling the deadlock. There might be
2077 a sleep-unlock race, but the locking primitive needs to resolve
2078 such races properly in any case.
2080 Locks and semaphores may not provide any guarantee of ordering on UP compiled
2081 systems, and so cannot be counted on in such a situation to actually achieve
2082 anything at all - especially with respect to I/O accesses - unless combined
2083 with interrupt disabling operations.
2085 See also the section on "Inter-CPU acquiring barrier effects".
2088 As an example, consider the following:
2090 *A = a;
2091 *B = b;
2092 ACQUIRE
2093 *C = c;
2094 *D = d;
2095 RELEASE
2096 *E = e;
2097 *F = f;
2099 The following sequence of events is acceptable:
2101 ACQUIRE, {*F,*A}, *E, {*C,*D}, *B, RELEASE
2103 [+] Note that {*F,*A} indicates a combined access.
2105 But none of the following are:
2107 {*F,*A}, *B, ACQUIRE, *C, *D, RELEASE, *E
2108 *A, *B, *C, ACQUIRE, *D, RELEASE, *E, *F
2109 *A, *B, ACQUIRE, *C, RELEASE, *D, *E, *F
2110 *B, ACQUIRE, *C, *D, RELEASE, {*F,*A}, *E
2114 INTERRUPT DISABLING FUNCTIONS
2115 -----------------------------
2117 Functions that disable interrupts (ACQUIRE equivalent) and enable interrupts
2118 (RELEASE equivalent) will act as compiler barriers only. So if memory or I/O
2119 barriers are required in such a situation, they must be provided from some
2120 other means.
2123 SLEEP AND WAKE-UP FUNCTIONS
2124 ---------------------------
2126 Sleeping and waking on an event flagged in global data can be viewed as an
2127 interaction between two pieces of data: the task state of the task waiting for
2128 the event and the global data used to indicate the event. To make sure that
2129 these appear to happen in the right order, the primitives to begin the process
2130 of going to sleep, and the primitives to initiate a wake up imply certain
2131 barriers.
2133 Firstly, the sleeper normally follows something like this sequence of events:
2135 for (;;) {
2136 set_current_state(TASK_UNINTERRUPTIBLE);
2137 if (event_indicated)
2138 break;
2139 schedule();
2140 }
2142 A general memory barrier is interpolated automatically by set_current_state()
2143 after it has altered the task state:
2145 CPU 1
2146 ===============================
2147 set_current_state();
2148 smp_store_mb();
2149 STORE current->state
2150 <general barrier>
2151 LOAD event_indicated
2153 set_current_state() may be wrapped by:
2155 prepare_to_wait();
2156 prepare_to_wait_exclusive();
2158 which therefore also imply a general memory barrier after setting the state.
2159 The whole sequence above is available in various canned forms, all of which
2160 interpolate the memory barrier in the right place:
2162 wait_event();
2163 wait_event_interruptible();
2164 wait_event_interruptible_exclusive();
2165 wait_event_interruptible_timeout();
2166 wait_event_killable();
2167 wait_event_timeout();
2168 wait_on_bit();
2169 wait_on_bit_lock();
2172 Secondly, code that performs a wake up normally follows something like this:
2174 event_indicated = 1;
2175 wake_up(&event_wait_queue);
2177 or:
2179 event_indicated = 1;
2180 wake_up_process(event_daemon);
2182 A general memory barrier is executed by wake_up() if it wakes something up.
2183 If it doesn't wake anything up then a memory barrier may or may not be
2184 executed; you must not rely on it. The barrier occurs before the task state
2185 is accessed, in particular, it sits between the STORE to indicate the event
2186 and the STORE to set TASK_RUNNING:
2188 CPU 1 (Sleeper) CPU 2 (Waker)
2189 =============================== ===============================
2190 set_current_state(); STORE event_indicated
2191 smp_store_mb(); wake_up();
2192 STORE current->state ...
2193 <general barrier> <general barrier>
2194 LOAD event_indicated if ((LOAD task->state) & TASK_NORMAL)
2195 STORE task->state
2197 where "task" is the thread being woken up and it equals CPU 1's "current".
2199 To repeat, a general memory barrier is guaranteed to be executed by wake_up()
2200 if something is actually awakened, but otherwise there is no such guarantee.
2201 To see this, consider the following sequence of events, where X and Y are both
2202 initially zero:
2204 CPU 1 CPU 2
2205 =============================== ===============================
2206 X = 1; Y = 1;
2207 smp_mb(); wake_up();
2208 LOAD Y LOAD X
2210 If a wakeup does occur, one (at least) of the two loads must see 1. If, on
2211 the other hand, a wakeup does not occur, both loads might see 0.
2213 wake_up_process() always executes a general memory barrier. The barrier again
2214 occurs before the task state is accessed. In particular, if the wake_up() in
2215 the previous snippet were replaced by a call to wake_up_process() then one of
2216 the two loads would be guaranteed to see 1.
2218 The available waker functions include:
2220 complete();
2221 wake_up();
2222 wake_up_all();
2223 wake_up_bit();
2224 wake_up_interruptible();
2225 wake_up_interruptible_all();
2226 wake_up_interruptible_nr();
2227 wake_up_interruptible_poll();
2228 wake_up_interruptible_sync();
2229 wake_up_interruptible_sync_poll();
2230 wake_up_locked();
2231 wake_up_locked_poll();
2232 wake_up_nr();
2233 wake_up_poll();
2234 wake_up_process();
2236 In terms of memory ordering, these functions all provide the same guarantees of
2237 a wake_up() (or stronger).
2239 [!] Note that the memory barriers implied by the sleeper and the waker do _not_
2240 order multiple stores before the wake-up with respect to loads of those stored
2241 values after the sleeper has called set_current_state(). For instance, if the
2242 sleeper does:
2244 set_current_state(TASK_INTERRUPTIBLE);
2245 if (event_indicated)
2246 break;
2247 __set_current_state(TASK_RUNNING);
2248 do_something(my_data);
2250 and the waker does:
2252 my_data = value;
2253 event_indicated = 1;
2254 wake_up(&event_wait_queue);
2256 there's no guarantee that the change to event_indicated will be perceived by
2257 the sleeper as coming after the change to my_data. In such a circumstance, the
2258 code on both sides must interpolate its own memory barriers between the
2259 separate data accesses. Thus the above sleeper ought to do:
2261 set_current_state(TASK_INTERRUPTIBLE);
2262 if (event_indicated) {
2263 smp_rmb();
2264 do_something(my_data);
2265 }
2267 and the waker should do:
2269 my_data = value;
2270 smp_wmb();
2271 event_indicated = 1;
2272 wake_up(&event_wait_queue);
2275 MISCELLANEOUS FUNCTIONS
2276 -----------------------
2278 Other functions that imply barriers:
2280 (*) schedule() and similar imply full memory barriers.
2283 ===================================
2284 INTER-CPU ACQUIRING BARRIER EFFECTS
2285 ===================================
2287 On SMP systems locking primitives give a more substantial form of barrier: one
2288 that does affect memory access ordering on other CPUs, within the context of
2289 conflict on any particular lock.
2292 ACQUIRES VS MEMORY ACCESSES
2293 ---------------------------
2295 Consider the following: the system has a pair of spinlocks (M) and (Q), and
2296 three CPUs; then should the following sequence of events occur:
2298 CPU 1 CPU 2
2299 =============================== ===============================
2300 WRITE_ONCE(*A, a); WRITE_ONCE(*E, e);
2301 ACQUIRE M ACQUIRE Q
2302 WRITE_ONCE(*B, b); WRITE_ONCE(*F, f);
2303 WRITE_ONCE(*C, c); WRITE_ONCE(*G, g);
2304 RELEASE M RELEASE Q
2305 WRITE_ONCE(*D, d); WRITE_ONCE(*H, h);
2307 Then there is no guarantee as to what order CPU 3 will see the accesses to *A
2308 through *H occur in, other than the constraints imposed by the separate locks
2309 on the separate CPUs. It might, for example, see:
2311 *E, ACQUIRE M, ACQUIRE Q, *G, *C, *F, *A, *B, RELEASE Q, *D, *H, RELEASE M
2313 But it won't see any of:
2315 *B, *C or *D preceding ACQUIRE M
2316 *A, *B or *C following RELEASE M
2317 *F, *G or *H preceding ACQUIRE Q
2318 *E, *F or *G following RELEASE Q
2322 ACQUIRES VS I/O ACCESSES
2323 ------------------------
2325 Under certain circumstances (especially involving NUMA), I/O accesses within
2326 two spinlocked sections on two different CPUs may be seen as interleaved by the
2327 PCI bridge, because the PCI bridge does not necessarily participate in the
2328 cache-coherence protocol, and is therefore incapable of issuing the required
2329 read memory barriers.
2331 For example:
2333 CPU 1 CPU 2
2334 =============================== ===============================
2335 spin_lock(Q)
2336 writel(0, ADDR)
2337 writel(1, DATA);
2338 spin_unlock(Q);
2339 spin_lock(Q);
2340 writel(4, ADDR);
2341 writel(5, DATA);
2342 spin_unlock(Q);
2344 may be seen by the PCI bridge as follows:
2346 STORE *ADDR = 0, STORE *ADDR = 4, STORE *DATA = 1, STORE *DATA = 5
2348 which would probably cause the hardware to malfunction.
2351 What is necessary here is to intervene with an mmiowb() before dropping the
2352 spinlock, for example:
2354 CPU 1 CPU 2
2355 =============================== ===============================
2356 spin_lock(Q)
2357 writel(0, ADDR)
2358 writel(1, DATA);
2359 mmiowb();
2360 spin_unlock(Q);
2361 spin_lock(Q);
2362 writel(4, ADDR);
2363 writel(5, DATA);
2364 mmiowb();
2365 spin_unlock(Q);
2367 this will ensure that the two stores issued on CPU 1 appear at the PCI bridge
2368 before either of the stores issued on CPU 2.
2371 Furthermore, following a store by a load from the same device obviates the need
2372 for the mmiowb(), because the load forces the store to complete before the load
2373 is performed:
2375 CPU 1 CPU 2
2376 =============================== ===============================
2377 spin_lock(Q)
2378 writel(0, ADDR)
2379 a = readl(DATA);
2380 spin_unlock(Q);
2381 spin_lock(Q);
2382 writel(4, ADDR);
2383 b = readl(DATA);
2384 spin_unlock(Q);
2387 See Documentation/driver-api/device-io.rst for more information.
2390 =================================
2391 WHERE ARE MEMORY BARRIERS NEEDED?
2392 =================================
2394 Under normal operation, memory operation reordering is generally not going to
2395 be a problem as a single-threaded linear piece of code will still appear to
2396 work correctly, even if it's in an SMP kernel. There are, however, four
2397 circumstances in which reordering definitely _could_ be a problem:
2399 (*) Interprocessor interaction.
2401 (*) Atomic operations.
2403 (*) Accessing devices.
2405 (*) Interrupts.
2408 INTERPROCESSOR INTERACTION
2409 --------------------------
2411 When there's a system with more than one processor, more than one CPU in the
2412 system may be working on the same data set at the same time. This can cause
2413 synchronisation problems, and the usual way of dealing with them is to use
2414 locks. Locks, however, are quite expensive, and so it may be preferable to
2415 operate without the use of a lock if at all possible. In such a case
2416 operations that affect both CPUs may have to be carefully ordered to prevent
2417 a malfunction.
2419 Consider, for example, the R/W semaphore slow path. Here a waiting process is
2420 queued on the semaphore, by virtue of it having a piece of its stack linked to
2421 the semaphore's list of waiting processes:
2423 struct rw_semaphore {
2424 ...
2425 spinlock_t lock;
2426 struct list_head waiters;
2427 };
2429 struct rwsem_waiter {
2430 struct list_head list;
2431 struct task_struct *task;
2432 };
2434 To wake up a particular waiter, the up_read() or up_write() functions have to:
2436 (1) read the next pointer from this waiter's record to know as to where the
2437 next waiter record is;
2439 (2) read the pointer to the waiter's task structure;
2441 (3) clear the task pointer to tell the waiter it has been given the semaphore;
2443 (4) call wake_up_process() on the task; and
2445 (5) release the reference held on the waiter's task struct.
2447 In other words, it has to perform this sequence of events:
2449 LOAD waiter->list.next;
2450 LOAD waiter->task;
2451 STORE waiter->task;
2452 CALL wakeup
2453 RELEASE task
2455 and if any of these steps occur out of order, then the whole thing may
2456 malfunction.
2458 Once it has queued itself and dropped the semaphore lock, the waiter does not
2459 get the lock again; it instead just waits for its task pointer to be cleared
2460 before proceeding. Since the record is on the waiter's stack, this means that
2461 if the task pointer is cleared _before_ the next pointer in the list is read,
2462 another CPU might start processing the waiter and might clobber the waiter's
2463 stack before the up*() function has a chance to read the next pointer.
2465 Consider then what might happen to the above sequence of events:
2467 CPU 1 CPU 2
2468 =============================== ===============================
2469 down_xxx()
2470 Queue waiter
2471 Sleep
2472 up_yyy()
2473 LOAD waiter->task;
2474 STORE waiter->task;
2475 Woken up by other event
2476 <preempt>
2477 Resume processing
2478 down_xxx() returns
2479 call foo()
2480 foo() clobbers *waiter
2481 </preempt>
2482 LOAD waiter->list.next;
2483 --- OOPS ---
2485 This could be dealt with using the semaphore lock, but then the down_xxx()
2486 function has to needlessly get the spinlock again after being woken up.
2488 The way to deal with this is to insert a general SMP memory barrier:
2490 LOAD waiter->list.next;
2491 LOAD waiter->task;
2492 smp_mb();
2493 STORE waiter->task;
2494 CALL wakeup
2495 RELEASE task
2497 In this case, the barrier makes a guarantee that all memory accesses before the
2498 barrier will appear to happen before all the memory accesses after the barrier
2499 with respect to the other CPUs on the system. It does _not_ guarantee that all
2500 the memory accesses before the barrier will be complete by the time the barrier
2501 instruction itself is complete.
2503 On a UP system - where this wouldn't be a problem - the smp_mb() is just a
2504 compiler barrier, thus making sure the compiler emits the instructions in the
2505 right order without actually intervening in the CPU. Since there's only one
2506 CPU, that CPU's dependency ordering logic will take care of everything else.
2509 ATOMIC OPERATIONS
2510 -----------------
2512 Whilst they are technically interprocessor interaction considerations, atomic
2513 operations are noted specially as some of them imply full memory barriers and
2514 some don't, but they're very heavily relied on as a group throughout the
2515 kernel.
2517 See Documentation/atomic_t.txt for more information.
2520 ACCESSING DEVICES
2521 -----------------
2523 Many devices can be memory mapped, and so appear to the CPU as if they're just
2524 a set of memory locations. To control such a device, the driver usually has to
2525 make the right memory accesses in exactly the right order.
2527 However, having a clever CPU or a clever compiler creates a potential problem
2528 in that the carefully sequenced accesses in the driver code won't reach the
2529 device in the requisite order if the CPU or the compiler thinks it is more
2530 efficient to reorder, combine or merge accesses - something that would cause
2531 the device to malfunction.
2533 Inside of the Linux kernel, I/O should be done through the appropriate accessor
2534 routines - such as inb() or writel() - which know how to make such accesses
2535 appropriately sequential. Whilst this, for the most part, renders the explicit
2536 use of memory barriers unnecessary, there are a couple of situations where they
2537 might be needed:
2539 (1) On some systems, I/O stores are not strongly ordered across all CPUs, and
2540 so for _all_ general drivers locks should be used and mmiowb() must be
2541 issued prior to unlocking the critical section.
2543 (2) If the accessor functions are used to refer to an I/O memory window with
2544 relaxed memory access properties, then _mandatory_ memory barriers are
2545 required to enforce ordering.
2547 See Documentation/driver-api/device-io.rst for more information.
2550 INTERRUPTS
2551 ----------
2553 A driver may be interrupted by its own interrupt service routine, and thus the
2554 two parts of the driver may interfere with each other's attempts to control or
2555 access the device.
2557 This may be alleviated - at least in part - by disabling local interrupts (a
2558 form of locking), such that the critical operations are all contained within
2559 the interrupt-disabled section in the driver. Whilst the driver's interrupt
2560 routine is executing, the driver's core may not run on the same CPU, and its
2561 interrupt is not permitted to happen again until the current interrupt has been
2562 handled, thus the interrupt handler does not need to lock against that.
2564 However, consider a driver that was talking to an ethernet card that sports an
2565 address register and a data register. If that driver's core talks to the card
2566 under interrupt-disablement and then the driver's interrupt handler is invoked:
2568 LOCAL IRQ DISABLE
2569 writew(ADDR, 3);
2570 writew(DATA, y);
2571 LOCAL IRQ ENABLE
2572 <interrupt>
2573 writew(ADDR, 4);
2574 q = readw(DATA);
2575 </interrupt>
2577 The store to the data register might happen after the second store to the
2578 address register if ordering rules are sufficiently relaxed:
2580 STORE *ADDR = 3, STORE *ADDR = 4, STORE *DATA = y, q = LOAD *DATA
2583 If ordering rules are relaxed, it must be assumed that accesses done inside an
2584 interrupt disabled section may leak outside of it and may interleave with
2585 accesses performed in an interrupt - and vice versa - unless implicit or
2586 explicit barriers are used.
2588 Normally this won't be a problem because the I/O accesses done inside such
2589 sections will include synchronous load operations on strictly ordered I/O
2590 registers that form implicit I/O barriers. If this isn't sufficient then an
2591 mmiowb() may need to be used explicitly.
2594 A similar situation may occur between an interrupt routine and two routines
2595 running on separate CPUs that communicate with each other. If such a case is
2596 likely, then interrupt-disabling locks should be used to guarantee ordering.
2599 ==========================
2600 KERNEL I/O BARRIER EFFECTS
2601 ==========================
2603 When accessing I/O memory, drivers should use the appropriate accessor
2604 functions:
2606 (*) inX(), outX():
2608 These are intended to talk to I/O space rather than memory space, but
2609 that's primarily a CPU-specific concept. The i386 and x86_64 processors
2610 do indeed have special I/O space access cycles and instructions, but many
2611 CPUs don't have such a concept.
2613 The PCI bus, amongst others, defines an I/O space concept which - on such
2614 CPUs as i386 and x86_64 - readily maps to the CPU's concept of I/O
2615 space. However, it may also be mapped as a virtual I/O space in the CPU's
2616 memory map, particularly on those CPUs that don't support alternate I/O
2617 spaces.
2619 Accesses to this space may be fully synchronous (as on i386), but
2620 intermediary bridges (such as the PCI host bridge) may not fully honour
2621 that.
2623 They are guaranteed to be fully ordered with respect to each other.
2625 They are not guaranteed to be fully ordered with respect to other types of
2626 memory and I/O operation.
2628 (*) readX(), writeX():
2630 Whether these are guaranteed to be fully ordered and uncombined with
2631 respect to each other on the issuing CPU depends on the characteristics
2632 defined for the memory window through which they're accessing. On later
2633 i386 architecture machines, for example, this is controlled by way of the
2634 MTRR registers.
2636 Ordinarily, these will be guaranteed to be fully ordered and uncombined,
2637 provided they're not accessing a prefetchable device.
2639 However, intermediary hardware (such as a PCI bridge) may indulge in
2640 deferral if it so wishes; to flush a store, a load from the same location
2641 is preferred[*], but a load from the same device or from configuration
2642 space should suffice for PCI.
2644 [*] NOTE! attempting to load from the same location as was written to may
2645 cause a malfunction - consider the 16550 Rx/Tx serial registers for
2646 example.
2648 Used with prefetchable I/O memory, an mmiowb() barrier may be required to
2649 force stores to be ordered.
2651 Please refer to the PCI specification for more information on interactions
2652 between PCI transactions.
2654 (*) readX_relaxed(), writeX_relaxed()
2656 These are similar to readX() and writeX(), but provide weaker memory
2657 ordering guarantees. Specifically, they do not guarantee ordering with
2658 respect to normal memory accesses (e.g. DMA buffers) nor do they guarantee
2659 ordering with respect to LOCK or UNLOCK operations. If the latter is
2660 required, an mmiowb() barrier can be used. Note that relaxed accesses to
2661 the same peripheral are guaranteed to be ordered with respect to each
2662 other.
2664 (*) ioreadX(), iowriteX()
2666 These will perform appropriately for the type of access they're actually
2667 doing, be it inX()/outX() or readX()/writeX().
2670 ========================================
2671 ASSUMED MINIMUM EXECUTION ORDERING MODEL
2672 ========================================
2674 It has to be assumed that the conceptual CPU is weakly-ordered but that it will
2675 maintain the appearance of program causality with respect to itself. Some CPUs
2676 (such as i386 or x86_64) are more constrained than others (such as powerpc or
2677 frv), and so the most relaxed case (namely DEC Alpha) must be assumed outside
2678 of arch-specific code.
2680 This means that it must be considered that the CPU will execute its instruction
2681 stream in any order it feels like - or even in parallel - provided that if an
2682 instruction in the stream depends on an earlier instruction, then that
2683 earlier instruction must be sufficiently complete[*] before the later
2684 instruction may proceed; in other words: provided that the appearance of
2685 causality is maintained.
2687 [*] Some instructions have more than one effect - such as changing the
2688 condition codes, changing registers or changing memory - and different
2689 instructions may depend on different effects.
2691 A CPU may also discard any instruction sequence that winds up having no
2692 ultimate effect. For example, if two adjacent instructions both load an
2693 immediate value into the same register, the first may be discarded.
2696 Similarly, it has to be assumed that compiler might reorder the instruction
2697 stream in any way it sees fit, again provided the appearance of causality is
2698 maintained.
2701 ============================
2702 THE EFFECTS OF THE CPU CACHE
2703 ============================
2705 The way cached memory operations are perceived across the system is affected to
2706 a certain extent by the caches that lie between CPUs and memory, and by the
2707 memory coherence system that maintains the consistency of state in the system.
2709 As far as the way a CPU interacts with another part of the system through the
2710 caches goes, the memory system has to include the CPU's caches, and memory
2711 barriers for the most part act at the interface between the CPU and its cache
2712 (memory barriers logically act on the dotted line in the following diagram):
2714 <--- CPU ---> : <----------- Memory ----------->
2715 :
2716 +--------+ +--------+ : +--------+ +-----------+
2717 | | | | : | | | | +--------+
2718 | CPU | | Memory | : | CPU | | | | |
2719 | Core |--->| Access |----->| Cache |<-->| | | |
2720 | | | Queue | : | | | |--->| Memory |
2721 | | | | : | | | | | |
2722 +--------+ +--------+ : +--------+ | | | |
2723 : | Cache | +--------+
2724 : | Coherency |
2725 : | Mechanism | +--------+
2726 +--------+ +--------+ : +--------+ | | | |
2727 | | | | : | | | | | |
2728 | CPU | | Memory | : | CPU | | |--->| Device |
2729 | Core |--->| Access |----->| Cache |<-->| | | |
2730 | | | Queue | : | | | | | |
2731 | | | | : | | | | +--------+
2732 +--------+ +--------+ : +--------+ +-----------+
2733 :
2734 :
2736 Although any particular load or store may not actually appear outside of the
2737 CPU that issued it since it may have been satisfied within the CPU's own cache,
2738 it will still appear as if the full memory access had taken place as far as the
2739 other CPUs are concerned since the cache coherency mechanisms will migrate the
2740 cacheline over to the accessing CPU and propagate the effects upon conflict.
2742 The CPU core may execute instructions in any order it deems fit, provided the
2743 expected program causality appears to be maintained. Some of the instructions
2744 generate load and store operations which then go into the queue of memory
2745 accesses to be performed. The core may place these in the queue in any order
2746 it wishes, and continue execution until it is forced to wait for an instruction
2747 to complete.
2749 What memory barriers are concerned with is controlling the order in which
2750 accesses cross from the CPU side of things to the memory side of things, and
2751 the order in which the effects are perceived to happen by the other observers
2752 in the system.
2754 [!] Memory barriers are _not_ needed within a given CPU, as CPUs always see
2755 their own loads and stores as if they had happened in program order.
2757 [!] MMIO or other device accesses may bypass the cache system. This depends on
2758 the properties of the memory window through which devices are accessed and/or
2759 the use of any special device communication instructions the CPU may have.
2762 CACHE COHERENCY
2763 ---------------
2765 Life isn't quite as simple as it may appear above, however: for while the
2766 caches are expected to be coherent, there's no guarantee that that coherency
2767 will be ordered. This means that whilst changes made on one CPU will
2768 eventually become visible on all CPUs, there's no guarantee that they will
2769 become apparent in the same order on those other CPUs.
2772 Consider dealing with a system that has a pair of CPUs (1 & 2), each of which
2773 has a pair of parallel data caches (CPU 1 has A/B, and CPU 2 has C/D):
2775 :
2776 : +--------+
2777 : +---------+ | |
2778 +--------+ : +--->| Cache A |<------->| |
2779 | | : | +---------+ | |
2780 | CPU 1 |<---+ | |
2781 | | : | +---------+ | |
2782 +--------+ : +--->| Cache B |<------->| |
2783 : +---------+ | |
2784 : | Memory |
2785 : +---------+ | System |
2786 +--------+ : +--->| Cache C |<------->| |
2787 | | : | +---------+ | |
2788 | CPU 2 |<---+ | |
2789 | | : | +---------+ | |
2790 +--------+ : +--->| Cache D |<------->| |
2791 : +---------+ | |
2792 : +--------+
2793 :
2795 Imagine the system has the following properties:
2797 (*) an odd-numbered cache line may be in cache A, cache C or it may still be
2798 resident in memory;
2800 (*) an even-numbered cache line may be in cache B, cache D or it may still be
2801 resident in memory;
2803 (*) whilst the CPU core is interrogating one cache, the other cache may be
2804 making use of the bus to access the rest of the system - perhaps to
2805 displace a dirty cacheline or to do a speculative load;
2807 (*) each cache has a queue of operations that need to be applied to that cache
2808 to maintain coherency with the rest of the system;
2810 (*) the coherency queue is not flushed by normal loads to lines already
2811 present in the cache, even though the contents of the queue may
2812 potentially affect those loads.
2814 Imagine, then, that two writes are made on the first CPU, with a write barrier
2815 between them to guarantee that they will appear to reach that CPU's caches in
2816 the requisite order:
2818 CPU 1 CPU 2 COMMENT
2819 =============== =============== =======================================
2820 u == 0, v == 1 and p == &u, q == &u
2821 v = 2;
2822 smp_wmb(); Make sure change to v is visible before
2823 change to p
2824 <A:modify v=2> v is now in cache A exclusively
2825 p = &v;
2826 <B:modify p=&v> p is now in cache B exclusively
2828 The write memory barrier forces the other CPUs in the system to perceive that
2829 the local CPU's caches have apparently been updated in the correct order. But
2830 now imagine that the second CPU wants to read those values:
2832 CPU 1 CPU 2 COMMENT
2833 =============== =============== =======================================
2834 ...
2835 q = p;
2836 x = *q;
2838 The above pair of reads may then fail to happen in the expected order, as the
2839 cacheline holding p may get updated in one of the second CPU's caches whilst
2840 the update to the cacheline holding v is delayed in the other of the second
2841 CPU's caches by some other cache event:
2843 CPU 1 CPU 2 COMMENT
2844 =============== =============== =======================================
2845 u == 0, v == 1 and p == &u, q == &u
2846 v = 2;
2847 smp_wmb();
2848 <A:modify v=2> <C:busy>
2849 <C:queue v=2>
2850 p = &v; q = p;
2851 <D:request p>
2852 <B:modify p=&v> <D:commit p=&v>
2853 <D:read p>
2854 x = *q;
2855 <C:read *q> Reads from v before v updated in cache
2856 <C:unbusy>
2857 <C:commit v=2>
2859 Basically, whilst both cachelines will be updated on CPU 2 eventually, there's
2860 no guarantee that, without intervention, the order of update will be the same
2861 as that committed on CPU 1.
2864 To intervene, we need to interpolate a data dependency barrier or a read
2865 barrier between the loads (which as of v4.15 is supplied unconditionally
2866 by the READ_ONCE() macro). This will force the cache to commit its
2867 coherency queue before processing any further requests:
2869 CPU 1 CPU 2 COMMENT
2870 =============== =============== =======================================
2871 u == 0, v == 1 and p == &u, q == &u
2872 v = 2;
2873 smp_wmb();
2874 <A:modify v=2> <C:busy>
2875 <C:queue v=2>
2876 p = &v; q = p;
2877 <D:request p>
2878 <B:modify p=&v> <D:commit p=&v>
2879 <D:read p>
2880 smp_read_barrier_depends()
2881 <C:unbusy>
2882 <C:commit v=2>
2883 x = *q;
2884 <C:read *q> Reads from v after v updated in cache
2887 This sort of problem can be encountered on DEC Alpha processors as they have a
2888 split cache that improves performance by making better use of the data bus.
2889 Whilst most CPUs do imply a data dependency barrier on the read when a memory
2890 access depends on a read, not all do, so it may not be relied on.
2892 Other CPUs may also have split caches, but must coordinate between the various
2893 cachelets for normal memory accesses. The semantics of the Alpha removes the
2894 need for hardware coordination in the absence of memory barriers, which
2895 permitted Alpha to sport higher CPU clock rates back in the day. However,
2896 please note that (again, as of v4.15) smp_read_barrier_depends() should not
2897 be used except in Alpha arch-specific code and within the READ_ONCE() macro.
2900 CACHE COHERENCY VS DMA
2901 ----------------------
2903 Not all systems maintain cache coherency with respect to devices doing DMA. In
2904 such cases, a device attempting DMA may obtain stale data from RAM because
2905 dirty cache lines may be resident in the caches of various CPUs, and may not
2906 have been written back to RAM yet. To deal with this, the appropriate part of
2907 the kernel must flush the overlapping bits of cache on each CPU (and maybe
2908 invalidate them as well).
2910 In addition, the data DMA'd to RAM by a device may be overwritten by dirty
2911 cache lines being written back to RAM from a CPU's cache after the device has
2912 installed its own data, or cache lines present in the CPU's cache may simply
2913 obscure the fact that RAM has been updated, until at such time as the cacheline
2914 is discarded from the CPU's cache and reloaded. To deal with this, the
2915 appropriate part of the kernel must invalidate the overlapping bits of the
2916 cache on each CPU.
2918 See Documentation/core-api/cachetlb.rst for more information on cache management.
2921 CACHE COHERENCY VS MMIO
2922 -----------------------
2924 Memory mapped I/O usually takes place through memory locations that are part of
2925 a window in the CPU's memory space that has different properties assigned than
2926 the usual RAM directed window.
2928 Amongst these properties is usually the fact that such accesses bypass the
2929 caching entirely and go directly to the device buses. This means MMIO accesses
2930 may, in effect, overtake accesses to cached memory that were emitted earlier.
2931 A memory barrier isn't sufficient in such a case, but rather the cache must be
2932 flushed between the cached memory write and the MMIO access if the two are in
2933 any way dependent.
2936 =========================
2937 THE THINGS CPUS GET UP TO
2938 =========================
2940 A programmer might take it for granted that the CPU will perform memory
2941 operations in exactly the order specified, so that if the CPU is, for example,
2942 given the following piece of code to execute:
2944 a = READ_ONCE(*A);
2945 WRITE_ONCE(*B, b);
2946 c = READ_ONCE(*C);
2947 d = READ_ONCE(*D);
2948 WRITE_ONCE(*E, e);
2950 they would then expect that the CPU will complete the memory operation for each
2951 instruction before moving on to the next one, leading to a definite sequence of
2952 operations as seen by external observers in the system:
2954 LOAD *A, STORE *B, LOAD *C, LOAD *D, STORE *E.
2957 Reality is, of course, much messier. With many CPUs and compilers, the above
2958 assumption doesn't hold because:
2960 (*) loads are more likely to need to be completed immediately to permit
2961 execution progress, whereas stores can often be deferred without a
2962 problem;
2964 (*) loads may be done speculatively, and the result discarded should it prove
2965 to have been unnecessary;
2967 (*) loads may be done speculatively, leading to the result having been fetched
2968 at the wrong time in the expected sequence of events;
2970 (*) the order of the memory accesses may be rearranged to promote better use
2971 of the CPU buses and caches;
2973 (*) loads and stores may be combined to improve performance when talking to
2974 memory or I/O hardware that can do batched accesses of adjacent locations,
2975 thus cutting down on transaction setup costs (memory and PCI devices may
2976 both be able to do this); and
2978 (*) the CPU's data cache may affect the ordering, and whilst cache-coherency
2979 mechanisms may alleviate this - once the store has actually hit the cache
2980 - there's no guarantee that the coherency management will be propagated in
2981 order to other CPUs.
2983 So what another CPU, say, might actually observe from the above piece of code
2984 is:
2986 LOAD *A, ..., LOAD {*C,*D}, STORE *E, STORE *B
2988 (Where "LOAD {*C,*D}" is a combined load)
2991 However, it is guaranteed that a CPU will be self-consistent: it will see its
2992 _own_ accesses appear to be correctly ordered, without the need for a memory
2993 barrier. For instance with the following code:
2995 U = READ_ONCE(*A);
2996 WRITE_ONCE(*A, V);
2997 WRITE_ONCE(*A, W);
2998 X = READ_ONCE(*A);
2999 WRITE_ONCE(*A, Y);
3000 Z = READ_ONCE(*A);
3002 and assuming no intervention by an external influence, it can be assumed that
3003 the final result will appear to be:
3005 U == the original value of *A
3006 X == W
3007 Z == Y
3008 *A == Y
3010 The code above may cause the CPU to generate the full sequence of memory
3011 accesses:
3013 U=LOAD *A, STORE *A=V, STORE *A=W, X=LOAD *A, STORE *A=Y, Z=LOAD *A
3015 in that order, but, without intervention, the sequence may have almost any
3016 combination of elements combined or discarded, provided the program's view
3017 of the world remains consistent. Note that READ_ONCE() and WRITE_ONCE()
3018 are -not- optional in the above example, as there are architectures
3019 where a given CPU might reorder successive loads to the same location.
3020 On such architectures, READ_ONCE() and WRITE_ONCE() do whatever is
3021 necessary to prevent this, for example, on Itanium the volatile casts
3022 used by READ_ONCE() and WRITE_ONCE() cause GCC to emit the special ld.acq
3023 and st.rel instructions (respectively) that prevent such reordering.
3025 The compiler may also combine, discard or defer elements of the sequence before
3026 the CPU even sees them.
3028 For instance:
3030 *A = V;
3031 *A = W;
3033 may be reduced to:
3035 *A = W;
3037 since, without either a write barrier or an WRITE_ONCE(), it can be
3038 assumed that the effect of the storage of V to *A is lost. Similarly:
3040 *A = Y;
3041 Z = *A;
3043 may, without a memory barrier or an READ_ONCE() and WRITE_ONCE(), be
3044 reduced to:
3046 *A = Y;
3047 Z = Y;
3049 and the LOAD operation never appear outside of the CPU.
3052 AND THEN THERE'S THE ALPHA
3053 --------------------------
3055 The DEC Alpha CPU is one of the most relaxed CPUs there is. Not only that,
3056 some versions of the Alpha CPU have a split data cache, permitting them to have
3057 two semantically-related cache lines updated at separate times. This is where
3058 the data dependency barrier really becomes necessary as this synchronises both
3059 caches with the memory coherence system, thus making it seem like pointer
3060 changes vs new data occur in the right order.
3062 The Alpha defines the Linux kernel's memory model, although as of v4.15
3063 the Linux kernel's addition of smp_read_barrier_depends() to READ_ONCE()
3064 greatly reduced Alpha's impact on the memory model.
3066 See the subsection on "Cache Coherency" above.
3069 VIRTUAL MACHINE GUESTS
3070 ----------------------
3072 Guests running within virtual machines might be affected by SMP effects even if
3073 the guest itself is compiled without SMP support. This is an artifact of
3074 interfacing with an SMP host while running an UP kernel. Using mandatory
3075 barriers for this use-case would be possible but is often suboptimal.
3077 To handle this case optimally, low-level virt_mb() etc macros are available.
3078 These have the same effect as smp_mb() etc when SMP is enabled, but generate
3079 identical code for SMP and non-SMP systems. For example, virtual machine guests
3080 should use virt_mb() rather than smp_mb() when synchronizing against a
3081 (possibly SMP) host.
3083 These are equivalent to smp_mb() etc counterparts in all other respects,
3084 in particular, they do not control MMIO effects: to control
3085 MMIO effects, use mandatory barriers.
3088 ============
3089 EXAMPLE USES
3090 ============
3092 CIRCULAR BUFFERS
3093 ----------------
3095 Memory barriers can be used to implement circular buffering without the need
3096 of a lock to serialise the producer with the consumer. See:
3098 Documentation/core-api/circular-buffers.rst
3100 for details.
3103 ==========
3104 REFERENCES
3105 ==========
3107 Alpha AXP Architecture Reference Manual, Second Edition (Sites & Witek,
3108 Digital Press)
3109 Chapter 5.2: Physical Address Space Characteristics
3110 Chapter 5.4: Caches and Write Buffers
3111 Chapter 5.5: Data Sharing
3112 Chapter 5.6: Read/Write Ordering
3114 AMD64 Architecture Programmer's Manual Volume 2: System Programming
3115 Chapter 7.1: Memory-Access Ordering
3116 Chapter 7.4: Buffering and Combining Memory Writes
3118 ARM Architecture Reference Manual (ARMv8, for ARMv8-A architecture profile)
3119 Chapter B2: The AArch64 Application Level Memory Model
3121 IA-32 Intel Architecture Software Developer's Manual, Volume 3:
3122 System Programming Guide
3123 Chapter 7.1: Locked Atomic Operations
3124 Chapter 7.2: Memory Ordering
3125 Chapter 7.4: Serializing Instructions
3127 The SPARC Architecture Manual, Version 9
3128 Chapter 8: Memory Models
3129 Appendix D: Formal Specification of the Memory Models
3130 Appendix J: Programming with the Memory Models
3132 Storage in the PowerPC (Stone and Fitzgerald)
3134 UltraSPARC Programmer Reference Manual
3135 Chapter 5: Memory Accesses and Cacheability
3136 Chapter 15: Sparc-V9 Memory Models
3138 UltraSPARC III Cu User's Manual
3139 Chapter 9: Memory Models
3141 UltraSPARC IIIi Processor User's Manual
3142 Chapter 8: Memory Models
3144 UltraSPARC Architecture 2005
3145 Chapter 9: Memory
3146 Appendix D: Formal Specifications of the Memory Models
3148 UltraSPARC T1 Supplement to the UltraSPARC Architecture 2005
3149 Chapter 8: Memory Models
3150 Appendix F: Caches and Cache Coherency
3152 Solaris Internals, Core Kernel Architecture, p63-68:
3153 Chapter 3.3: Hardware Considerations for Locks and
3154 Synchronization
3156 Unix Systems for Modern Architectures, Symmetric Multiprocessing and Caching
3157 for Kernel Programmers:
3158 Chapter 13: Other Memory Models
3160 Intel Itanium Architecture Software Developer's Manual: Volume 1:
3161 Section 2.6: Speculation
3162 Section 4.4: Memory Access