xref: /openbmc/linux/Documentation/atomic_t.txt (revision ad4455c6)
1
2On atomic types (atomic_t atomic64_t and atomic_long_t).
3
4The atomic type provides an interface to the architecture's means of atomic
5RMW operations between CPUs (atomic operations on MMIO are not supported and
6can lead to fatal traps on some platforms).
7
8API
9---
10
11The 'full' API consists of (atomic64_ and atomic_long_ prefixes omitted for
12brevity):
13
14Non-RMW ops:
15
16  atomic_read(), atomic_set()
17  atomic_read_acquire(), atomic_set_release()
18
19
20RMW atomic operations:
21
22Arithmetic:
23
24  atomic_{add,sub,inc,dec}()
25  atomic_{add,sub,inc,dec}_return{,_relaxed,_acquire,_release}()
26  atomic_fetch_{add,sub,inc,dec}{,_relaxed,_acquire,_release}()
27
28
29Bitwise:
30
31  atomic_{and,or,xor,andnot}()
32  atomic_fetch_{and,or,xor,andnot}{,_relaxed,_acquire,_release}()
33
34
35Swap:
36
37  atomic_xchg{,_relaxed,_acquire,_release}()
38  atomic_cmpxchg{,_relaxed,_acquire,_release}()
39  atomic_try_cmpxchg{,_relaxed,_acquire,_release}()
40
41
42Reference count (but please see refcount_t):
43
44  atomic_add_unless(), atomic_inc_not_zero()
45  atomic_sub_and_test(), atomic_dec_and_test()
46
47
48Misc:
49
50  atomic_inc_and_test(), atomic_add_negative()
51  atomic_dec_unless_positive(), atomic_inc_unless_negative()
52
53
54Barriers:
55
56  smp_mb__{before,after}_atomic()
57
58
59TYPES (signed vs unsigned)
60-----
61
62While atomic_t, atomic_long_t and atomic64_t use int, long and s64
63respectively (for hysterical raisins), the kernel uses -fno-strict-overflow
64(which implies -fwrapv) and defines signed overflow to behave like
652s-complement.
66
67Therefore, an explicitly unsigned variant of the atomic ops is strictly
68unnecessary and we can simply cast, there is no UB.
69
70There was a bug in UBSAN prior to GCC-8 that would generate UB warnings for
71signed types.
72
73With this we also conform to the C/C++ _Atomic behaviour and things like
74P1236R1.
75
76
77SEMANTICS
78---------
79
80Non-RMW ops:
81
82The non-RMW ops are (typically) regular LOADs and STOREs and are canonically
83implemented using READ_ONCE(), WRITE_ONCE(), smp_load_acquire() and
84smp_store_release() respectively. Therefore, if you find yourself only using
85the Non-RMW operations of atomic_t, you do not in fact need atomic_t at all
86and are doing it wrong.
87
88A note for the implementation of atomic_set{}() is that it must not break the
89atomicity of the RMW ops. That is:
90
91  C Atomic-RMW-ops-are-atomic-WRT-atomic_set
92
93  {
94    atomic_t v = ATOMIC_INIT(1);
95  }
96
97  P0(atomic_t *v)
98  {
99    (void)atomic_add_unless(v, 1, 0);
100  }
101
102  P1(atomic_t *v)
103  {
104    atomic_set(v, 0);
105  }
106
107  exists
108  (v=2)
109
110In this case we would expect the atomic_set() from CPU1 to either happen
111before the atomic_add_unless(), in which case that latter one would no-op, or
112_after_ in which case we'd overwrite its result. In no case is "2" a valid
113outcome.
114
115This is typically true on 'normal' platforms, where a regular competing STORE
116will invalidate a LL/SC or fail a CMPXCHG.
117
118The obvious case where this is not so is when we need to implement atomic ops
119with a lock:
120
121  CPU0						CPU1
122
123  atomic_add_unless(v, 1, 0);
124    lock();
125    ret = READ_ONCE(v->counter); // == 1
126						atomic_set(v, 0);
127    if (ret != u)				  WRITE_ONCE(v->counter, 0);
128      WRITE_ONCE(v->counter, ret + 1);
129    unlock();
130
131the typical solution is to then implement atomic_set{}() with atomic_xchg().
132
133
134RMW ops:
135
136These come in various forms:
137
138 - plain operations without return value: atomic_{}()
139
140 - operations which return the modified value: atomic_{}_return()
141
142   these are limited to the arithmetic operations because those are
143   reversible. Bitops are irreversible and therefore the modified value
144   is of dubious utility.
145
146 - operations which return the original value: atomic_fetch_{}()
147
148 - swap operations: xchg(), cmpxchg() and try_cmpxchg()
149
150 - misc; the special purpose operations that are commonly used and would,
151   given the interface, normally be implemented using (try_)cmpxchg loops but
152   are time critical and can, (typically) on LL/SC architectures, be more
153   efficiently implemented.
154
155All these operations are SMP atomic; that is, the operations (for a single
156atomic variable) can be fully ordered and no intermediate state is lost or
157visible.
158
159
160ORDERING  (go read memory-barriers.txt first)
161--------
162
163The rule of thumb:
164
165 - non-RMW operations are unordered;
166
167 - RMW operations that have no return value are unordered;
168
169 - RMW operations that have a return value are fully ordered;
170
171 - RMW operations that are conditional are unordered on FAILURE,
172   otherwise the above rules apply.
173
174Except of course when an operation has an explicit ordering like:
175
176 {}_relaxed: unordered
177 {}_acquire: the R of the RMW (or atomic_read) is an ACQUIRE
178 {}_release: the W of the RMW (or atomic_set)  is a  RELEASE
179
180Where 'unordered' is against other memory locations. Address dependencies are
181not defeated.
182
183Fully ordered primitives are ordered against everything prior and everything
184subsequent. Therefore a fully ordered primitive is like having an smp_mb()
185before and an smp_mb() after the primitive.
186
187
188The barriers:
189
190  smp_mb__{before,after}_atomic()
191
192only apply to the RMW atomic ops and can be used to augment/upgrade the
193ordering inherent to the op. These barriers act almost like a full smp_mb():
194smp_mb__before_atomic() orders all earlier accesses against the RMW op
195itself and all accesses following it, and smp_mb__after_atomic() orders all
196later accesses against the RMW op and all accesses preceding it. However,
197accesses between the smp_mb__{before,after}_atomic() and the RMW op are not
198ordered, so it is advisable to place the barrier right next to the RMW atomic
199op whenever possible.
200
201These helper barriers exist because architectures have varying implicit
202ordering on their SMP atomic primitives. For example our TSO architectures
203provide full ordered atomics and these barriers are no-ops.
204
205NOTE: when the atomic RmW ops are fully ordered, they should also imply a
206compiler barrier.
207
208Thus:
209
210  atomic_fetch_add();
211
212is equivalent to:
213
214  smp_mb__before_atomic();
215  atomic_fetch_add_relaxed();
216  smp_mb__after_atomic();
217
218However the atomic_fetch_add() might be implemented more efficiently.
219
220Further, while something like:
221
222  smp_mb__before_atomic();
223  atomic_dec(&X);
224
225is a 'typical' RELEASE pattern, the barrier is strictly stronger than
226a RELEASE because it orders preceding instructions against both the read
227and write parts of the atomic_dec(), and against all following instructions
228as well. Similarly, something like:
229
230  atomic_inc(&X);
231  smp_mb__after_atomic();
232
233is an ACQUIRE pattern (though very much not typical), but again the barrier is
234strictly stronger than ACQUIRE. As illustrated:
235
236  C Atomic-RMW+mb__after_atomic-is-stronger-than-acquire
237
238  {
239  }
240
241  P0(int *x, atomic_t *y)
242  {
243    r0 = READ_ONCE(*x);
244    smp_rmb();
245    r1 = atomic_read(y);
246  }
247
248  P1(int *x, atomic_t *y)
249  {
250    atomic_inc(y);
251    smp_mb__after_atomic();
252    WRITE_ONCE(*x, 1);
253  }
254
255  exists
256  (0:r0=1 /\ 0:r1=0)
257
258This should not happen; but a hypothetical atomic_inc_acquire() --
259(void)atomic_fetch_inc_acquire() for instance -- would allow the outcome,
260because it would not order the W part of the RMW against the following
261WRITE_ONCE.  Thus:
262
263  P0			P1
264
265			t = LL.acq *y (0)
266			t++;
267			*x = 1;
268  r0 = *x (1)
269  RMB
270  r1 = *y (0)
271			SC *y, t;
272
273is allowed.
274
275
276CMPXCHG vs TRY_CMPXCHG
277----------------------
278
279  int atomic_cmpxchg(atomic_t *ptr, int old, int new);
280  bool atomic_try_cmpxchg(atomic_t *ptr, int *oldp, int new);
281
282Both provide the same functionality, but try_cmpxchg() can lead to more
283compact code. The functions relate like:
284
285  bool atomic_try_cmpxchg(atomic_t *ptr, int *oldp, int new)
286  {
287    int ret, old = *oldp;
288    ret = atomic_cmpxchg(ptr, old, new);
289    if (ret != old)
290      *oldp = ret;
291    return ret == old;
292  }
293
294and:
295
296  int atomic_cmpxchg(atomic_t *ptr, int old, int new)
297  {
298    (void)atomic_try_cmpxchg(ptr, &old, new);
299    return old;
300  }
301
302Usage:
303
304  old = atomic_read(&v);			old = atomic_read(&v);
305  for (;;) {					do {
306    new = func(old);				  new = func(old);
307    tmp = atomic_cmpxchg(&v, old, new);		} while (!atomic_try_cmpxchg(&v, &old, new));
308    if (tmp == old)
309      break;
310    old = tmp;
311  }
312
313NB. try_cmpxchg() also generates better code on some platforms (notably x86)
314where the function more closely matches the hardware instruction.
315
316
317FORWARD PROGRESS
318----------------
319
320In general strong forward progress is expected of all unconditional atomic
321operations -- those in the Arithmetic and Bitwise classes and xchg(). However
322a fair amount of code also requires forward progress from the conditional
323atomic operations.
324
325Specifically 'simple' cmpxchg() loops are expected to not starve one another
326indefinitely. However, this is not evident on LL/SC architectures, because
327while an LL/SC architecture 'can/should/must' provide forward progress
328guarantees between competing LL/SC sections, such a guarantee does not
329transfer to cmpxchg() implemented using LL/SC. Consider:
330
331  old = atomic_read(&v);
332  do {
333    new = func(old);
334  } while (!atomic_try_cmpxchg(&v, &old, new));
335
336which on LL/SC becomes something like:
337
338  old = atomic_read(&v);
339  do {
340    new = func(old);
341  } while (!({
342    volatile asm ("1: LL  %[oldval], %[v]\n"
343                  "   CMP %[oldval], %[old]\n"
344                  "   BNE 2f\n"
345                  "   SC  %[new], %[v]\n"
346                  "   BNE 1b\n"
347                  "2:\n"
348                  : [oldval] "=&r" (oldval), [v] "m" (v)
349		  : [old] "r" (old), [new] "r" (new)
350                  : "memory");
351    success = (oldval == old);
352    if (!success)
353      old = oldval;
354    success; }));
355
356However, even the forward branch from the failed compare can cause the LL/SC
357to fail on some architectures, let alone whatever the compiler makes of the C
358loop body. As a result there is no guarantee what so ever the cacheline
359containing @v will stay on the local CPU and progress is made.
360
361Even native CAS architectures can fail to provide forward progress for their
362primitive (See Sparc64 for an example).
363
364Such implementations are strongly encouraged to add exponential backoff loops
365to a failed CAS in order to ensure some progress. Affected architectures are
366also strongly encouraged to inspect/audit the atomic fallbacks, refcount_t and
367their locking primitives.
368