xref: /openbmc/linux/lib/errseq.c (revision fb960bd2)
1 // SPDX-License-Identifier: GPL-2.0
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 /**
45  * errseq_set - set a errseq_t for later reporting
46  * @eseq: errseq_t field that should be set
47  * @err: error to set (must be between -1 and -MAX_ERRNO)
48  *
49  * This function sets the error in *eseq, and increments the sequence counter
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  *
54  * We do return the latest value here, primarily for debugging purposes. The
55  * return value should not be used as a previously sampled value in later calls
56  * as it will not have the SEEN flag set.
57  */
58 errseq_t errseq_set(errseq_t *eseq, int err)
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 }
108 EXPORT_SYMBOL(errseq_set);
109 
110 /**
111  * errseq_sample - grab current errseq_t value
112  * @eseq: pointer to errseq_t to be sampled
113  *
114  * This function allows callers to sample an errseq_t value, marking it as
115  * "seen" if required.
116  */
117 errseq_t errseq_sample(errseq_t *eseq)
118 {
119 	errseq_t old = READ_ONCE(*eseq);
120 	errseq_t new = old;
121 
122 	/*
123 	 * For the common case of no errors ever having been set, we can skip
124 	 * marking the SEEN bit. Once an error has been set, the value will
125 	 * never go back to zero.
126 	 */
127 	if (old != 0) {
128 		new |= ERRSEQ_SEEN;
129 		if (old != new)
130 			cmpxchg(eseq, old, new);
131 	}
132 	return new;
133 }
134 EXPORT_SYMBOL(errseq_sample);
135 
136 /**
137  * errseq_check - has an error occurred since a particular sample point?
138  * @eseq: pointer to errseq_t value to be checked
139  * @since: previously-sampled errseq_t from which to check
140  *
141  * Grab the value that eseq points to, and see if it has changed "since"
142  * the given value was sampled. The "since" value is not advanced, so there
143  * is no need to mark the value as seen.
144  *
145  * Returns the latest error set in the errseq_t or 0 if it hasn't changed.
146  */
147 int errseq_check(errseq_t *eseq, errseq_t since)
148 {
149 	errseq_t cur = READ_ONCE(*eseq);
150 
151 	if (likely(cur == since))
152 		return 0;
153 	return -(cur & MAX_ERRNO);
154 }
155 EXPORT_SYMBOL(errseq_check);
156 
157 /**
158  * errseq_check_and_advance - check an errseq_t and advance to current value
159  * @eseq: pointer to value being checked and reported
160  * @since: pointer to previously-sampled errseq_t to check against and advance
161  *
162  * Grab the eseq value, and see whether it matches the value that "since"
163  * points to. If it does, then just return 0.
164  *
165  * If it doesn't, then the value has changed. Set the "seen" flag, and try to
166  * swap it into place as the new eseq value. Then, set that value as the new
167  * "since" value, and return whatever the error portion is set to.
168  *
169  * Note that no locking is provided here for concurrent updates to the "since"
170  * value. The caller must provide that if necessary. Because of this, callers
171  * may want to do a lockless errseq_check before taking the lock and calling
172  * this.
173  */
174 int errseq_check_and_advance(errseq_t *eseq, errseq_t *since)
175 {
176 	int err = 0;
177 	errseq_t old, new;
178 
179 	/*
180 	 * Most callers will want to use the inline wrapper to check this,
181 	 * so that the common case of no error is handled without needing
182 	 * to take the lock that protects the "since" value.
183 	 */
184 	old = READ_ONCE(*eseq);
185 	if (old != *since) {
186 		/*
187 		 * Set the flag and try to swap it into place if it has
188 		 * changed.
189 		 *
190 		 * We don't care about the outcome of the swap here. If the
191 		 * swap doesn't occur, then it has either been updated by a
192 		 * writer who is altering the value in some way (updating
193 		 * counter or resetting the error), or another reader who is
194 		 * just setting the "seen" flag. Either outcome is OK, and we
195 		 * can advance "since" and return an error based on what we
196 		 * have.
197 		 */
198 		new = old | ERRSEQ_SEEN;
199 		if (new != old)
200 			cmpxchg(eseq, old, new);
201 		*since = new;
202 		err = -(new & MAX_ERRNO);
203 	}
204 	return err;
205 }
206 EXPORT_SYMBOL(errseq_check_and_advance);
207