Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
84cbadad JL |
2 | #include <linux/err.h> |
3 | #include <linux/bug.h> | |
4 | #include <linux/atomic.h> | |
5 | #include <linux/errseq.h> | |
6 | ||
7 | /* | |
8 | * An errseq_t is a way of recording errors in one place, and allowing any | |
9 | * number of "subscribers" to tell whether it has changed since a previous | |
10 | * point where it was sampled. | |
11 | * | |
12 | * It's implemented as an unsigned 32-bit value. The low order bits are | |
13 | * designated to hold an error code (between 0 and -MAX_ERRNO). The upper bits | |
14 | * are used as a counter. This is done with atomics instead of locking so that | |
15 | * these functions can be called from any context. | |
16 | * | |
17 | * The general idea is for consumers to sample an errseq_t value. That value | |
18 | * can later be used to tell whether any new errors have occurred since that | |
19 | * sampling was done. | |
20 | * | |
21 | * Note that there is a risk of collisions if new errors are being recorded | |
22 | * frequently, since we have so few bits to use as a counter. | |
23 | * | |
24 | * To mitigate this, one bit is used as a flag to tell whether the value has | |
25 | * been sampled since a new value was recorded. That allows us to avoid bumping | |
26 | * the counter if no one has sampled it since the last time an error was | |
27 | * recorded. | |
28 | * | |
29 | * A new errseq_t should always be zeroed out. A errseq_t value of all zeroes | |
30 | * is the special (but common) case where there has never been an error. An all | |
31 | * zero value thus serves as the "epoch" if one wishes to know whether there | |
32 | * has ever been an error set since it was first initialized. | |
33 | */ | |
34 | ||
35 | /* The low bits are designated for error code (max of MAX_ERRNO) */ | |
36 | #define ERRSEQ_SHIFT ilog2(MAX_ERRNO + 1) | |
37 | ||
38 | /* This bit is used as a flag to indicate whether the value has been seen */ | |
39 | #define ERRSEQ_SEEN (1 << ERRSEQ_SHIFT) | |
40 | ||
41 | /* The lowest bit of the counter */ | |
42 | #define ERRSEQ_CTR_INC (1 << (ERRSEQ_SHIFT + 1)) | |
43 | ||
44 | /** | |
3acdfd28 | 45 | * errseq_set - set a errseq_t for later reporting |
84cbadad | 46 | * @eseq: errseq_t field that should be set |
3acdfd28 | 47 | * @err: error to set (must be between -1 and -MAX_ERRNO) |
84cbadad | 48 | * |
14ebc28e | 49 | * This function sets the error in @eseq, and increments the sequence counter |
84cbadad JL |
50 | * if the last sequence was sampled at some point in the past. |
51 | * | |
52 | * Any error set will always overwrite an existing error. | |
53 | * | |
14ebc28e MW |
54 | * Return: The previous value, primarily for debugging purposes. The |
55 | * return value should not be used as a previously sampled value in later | |
56 | * calls as it will not have the SEEN flag set. | |
84cbadad | 57 | */ |
3acdfd28 | 58 | errseq_t errseq_set(errseq_t *eseq, int err) |
84cbadad JL |
59 | { |
60 | errseq_t cur, old; | |
61 | ||
62 | /* MAX_ERRNO must be able to serve as a mask */ | |
63 | BUILD_BUG_ON_NOT_POWER_OF_2(MAX_ERRNO + 1); | |
64 | ||
65 | /* | |
66 | * Ensure the error code actually fits where we want it to go. If it | |
67 | * doesn't then just throw a warning and don't record anything. We | |
68 | * also don't accept zero here as that would effectively clear a | |
69 | * previous error. | |
70 | */ | |
71 | old = READ_ONCE(*eseq); | |
72 | ||
73 | if (WARN(unlikely(err == 0 || (unsigned int)-err > MAX_ERRNO), | |
74 | "err = %d\n", err)) | |
75 | return old; | |
76 | ||
77 | for (;;) { | |
78 | errseq_t new; | |
79 | ||
80 | /* Clear out error bits and set new error */ | |
81 | new = (old & ~(MAX_ERRNO|ERRSEQ_SEEN)) | -err; | |
82 | ||
83 | /* Only increment if someone has looked at it */ | |
84 | if (old & ERRSEQ_SEEN) | |
85 | new += ERRSEQ_CTR_INC; | |
86 | ||
87 | /* If there would be no change, then call it done */ | |
88 | if (new == old) { | |
89 | cur = new; | |
90 | break; | |
91 | } | |
92 | ||
93 | /* Try to swap the new value into place */ | |
94 | cur = cmpxchg(eseq, old, new); | |
95 | ||
96 | /* | |
97 | * Call it success if we did the swap or someone else beat us | |
98 | * to it for the same value. | |
99 | */ | |
100 | if (likely(cur == old || cur == new)) | |
101 | break; | |
102 | ||
103 | /* Raced with an update, try again */ | |
104 | old = cur; | |
105 | } | |
106 | return cur; | |
107 | } | |
3acdfd28 | 108 | EXPORT_SYMBOL(errseq_set); |
84cbadad JL |
109 | |
110 | /** | |
14ebc28e MW |
111 | * errseq_sample() - Grab current errseq_t value. |
112 | * @eseq: Pointer to errseq_t to be sampled. | |
84cbadad JL |
113 | * |
114 | * This function allows callers to sample an errseq_t value, marking it as | |
115 | * "seen" if required. | |
14ebc28e MW |
116 | * |
117 | * Return: The current errseq value. | |
84cbadad JL |
118 | */ |
119 | errseq_t errseq_sample(errseq_t *eseq) | |
120 | { | |
121 | errseq_t old = READ_ONCE(*eseq); | |
122 | errseq_t new = old; | |
123 | ||
124 | /* | |
125 | * For the common case of no errors ever having been set, we can skip | |
126 | * marking the SEEN bit. Once an error has been set, the value will | |
127 | * never go back to zero. | |
128 | */ | |
129 | if (old != 0) { | |
130 | new |= ERRSEQ_SEEN; | |
131 | if (old != new) | |
132 | cmpxchg(eseq, old, new); | |
133 | } | |
134 | return new; | |
135 | } | |
136 | EXPORT_SYMBOL(errseq_sample); | |
137 | ||
138 | /** | |
14ebc28e MW |
139 | * errseq_check() - Has an error occurred since a particular sample point? |
140 | * @eseq: Pointer to errseq_t value to be checked. | |
141 | * @since: Previously-sampled errseq_t from which to check. | |
84cbadad | 142 | * |
14ebc28e MW |
143 | * Grab the value that eseq points to, and see if it has changed @since |
144 | * the given value was sampled. The @since value is not advanced, so there | |
84cbadad JL |
145 | * is no need to mark the value as seen. |
146 | * | |
14ebc28e | 147 | * Return: The latest error set in the errseq_t or 0 if it hasn't changed. |
84cbadad JL |
148 | */ |
149 | int errseq_check(errseq_t *eseq, errseq_t since) | |
150 | { | |
151 | errseq_t cur = READ_ONCE(*eseq); | |
152 | ||
153 | if (likely(cur == since)) | |
154 | return 0; | |
155 | return -(cur & MAX_ERRNO); | |
156 | } | |
157 | EXPORT_SYMBOL(errseq_check); | |
158 | ||
159 | /** | |
14ebc28e MW |
160 | * errseq_check_and_advance() - Check an errseq_t and advance to current value. |
161 | * @eseq: Pointer to value being checked and reported. | |
162 | * @since: Pointer to previously-sampled errseq_t to check against and advance. | |
84cbadad | 163 | * |
14ebc28e | 164 | * Grab the eseq value, and see whether it matches the value that @since |
84cbadad JL |
165 | * points to. If it does, then just return 0. |
166 | * | |
167 | * If it doesn't, then the value has changed. Set the "seen" flag, and try to | |
168 | * swap it into place as the new eseq value. Then, set that value as the new | |
169 | * "since" value, and return whatever the error portion is set to. | |
170 | * | |
171 | * Note that no locking is provided here for concurrent updates to the "since" | |
172 | * value. The caller must provide that if necessary. Because of this, callers | |
173 | * may want to do a lockless errseq_check before taking the lock and calling | |
174 | * this. | |
14ebc28e MW |
175 | * |
176 | * Return: Negative errno if one has been stored, or 0 if no new error has | |
177 | * occurred. | |
84cbadad JL |
178 | */ |
179 | int errseq_check_and_advance(errseq_t *eseq, errseq_t *since) | |
180 | { | |
181 | int err = 0; | |
182 | errseq_t old, new; | |
183 | ||
184 | /* | |
185 | * Most callers will want to use the inline wrapper to check this, | |
186 | * so that the common case of no error is handled without needing | |
187 | * to take the lock that protects the "since" value. | |
188 | */ | |
189 | old = READ_ONCE(*eseq); | |
190 | if (old != *since) { | |
191 | /* | |
192 | * Set the flag and try to swap it into place if it has | |
193 | * changed. | |
194 | * | |
195 | * We don't care about the outcome of the swap here. If the | |
196 | * swap doesn't occur, then it has either been updated by a | |
197 | * writer who is altering the value in some way (updating | |
198 | * counter or resetting the error), or another reader who is | |
199 | * just setting the "seen" flag. Either outcome is OK, and we | |
200 | * can advance "since" and return an error based on what we | |
201 | * have. | |
202 | */ | |
203 | new = old | ERRSEQ_SEEN; | |
204 | if (new != old) | |
205 | cmpxchg(eseq, old, new); | |
206 | *since = new; | |
207 | err = -(new & MAX_ERRNO); | |
208 | } | |
209 | return err; | |
210 | } | |
211 | EXPORT_SYMBOL(errseq_check_and_advance); |