xref: /openbmc/linux/lib/errseq.c (revision 7aacf86b)
1 #include <linux/err.h>
2 #include <linux/bug.h>
3 #include <linux/atomic.h>
4 #include <linux/errseq.h>
5 
6 /*
7  * An errseq_t is a way of recording errors in one place, and allowing any
8  * number of "subscribers" to tell whether it has changed since a previous
9  * point where it was sampled.
10  *
11  * It's implemented as an unsigned 32-bit value. The low order bits are
12  * designated to hold an error code (between 0 and -MAX_ERRNO). The upper bits
13  * are used as a counter. This is done with atomics instead of locking so that
14  * these functions can be called from any context.
15  *
16  * The general idea is for consumers to sample an errseq_t value. That value
17  * can later be used to tell whether any new errors have occurred since that
18  * sampling was done.
19  *
20  * Note that there is a risk of collisions if new errors are being recorded
21  * frequently, since we have so few bits to use as a counter.
22  *
23  * To mitigate this, one bit is used as a flag to tell whether the value has
24  * been sampled since a new value was recorded. That allows us to avoid bumping
25  * the counter if no one has sampled it since the last time an error was
26  * recorded.
27  *
28  * A new errseq_t should always be zeroed out.  A errseq_t value of all zeroes
29  * is the special (but common) case where there has never been an error. An all
30  * zero value thus serves as the "epoch" if one wishes to know whether there
31  * has ever been an error set since it was first initialized.
32  */
33 
34 /* The low bits are designated for error code (max of MAX_ERRNO) */
35 #define ERRSEQ_SHIFT		ilog2(MAX_ERRNO + 1)
36 
37 /* This bit is used as a flag to indicate whether the value has been seen */
38 #define ERRSEQ_SEEN		(1 << ERRSEQ_SHIFT)
39 
40 /* The lowest bit of the counter */
41 #define ERRSEQ_CTR_INC		(1 << (ERRSEQ_SHIFT + 1))
42 
43 /**
44  * __errseq_set - set a errseq_t for later reporting
45  * @eseq: errseq_t field that should be set
46  * @err: error to set
47  *
48  * This function sets the error in *eseq, and increments the sequence counter
49  * if the last sequence was sampled at some point in the past.
50  *
51  * Any error set will always overwrite an existing error.
52  *
53  * Most callers will want to use the errseq_set inline wrapper to efficiently
54  * handle the common case where err is 0.
55  *
56  * We do return an errseq_t here, primarily for debugging purposes. The return
57  * value should not be used as a previously sampled value in later calls as it
58  * will not have the SEEN flag set.
59  */
60 errseq_t __errseq_set(errseq_t *eseq, int err)
61 {
62 	errseq_t cur, old;
63 
64 	/* MAX_ERRNO must be able to serve as a mask */
65 	BUILD_BUG_ON_NOT_POWER_OF_2(MAX_ERRNO + 1);
66 
67 	/*
68 	 * Ensure the error code actually fits where we want it to go. If it
69 	 * doesn't then just throw a warning and don't record anything. We
70 	 * also don't accept zero here as that would effectively clear a
71 	 * previous error.
72 	 */
73 	old = READ_ONCE(*eseq);
74 
75 	if (WARN(unlikely(err == 0 || (unsigned int)-err > MAX_ERRNO),
76 				"err = %d\n", err))
77 		return old;
78 
79 	for (;;) {
80 		errseq_t new;
81 
82 		/* Clear out error bits and set new error */
83 		new = (old & ~(MAX_ERRNO|ERRSEQ_SEEN)) | -err;
84 
85 		/* Only increment if someone has looked at it */
86 		if (old & ERRSEQ_SEEN)
87 			new += ERRSEQ_CTR_INC;
88 
89 		/* If there would be no change, then call it done */
90 		if (new == old) {
91 			cur = new;
92 			break;
93 		}
94 
95 		/* Try to swap the new value into place */
96 		cur = cmpxchg(eseq, old, new);
97 
98 		/*
99 		 * Call it success if we did the swap or someone else beat us
100 		 * to it for the same value.
101 		 */
102 		if (likely(cur == old || cur == new))
103 			break;
104 
105 		/* Raced with an update, try again */
106 		old = cur;
107 	}
108 	return cur;
109 }
110 EXPORT_SYMBOL(__errseq_set);
111 
112 /**
113  * errseq_sample - grab current errseq_t value
114  * @eseq: pointer to errseq_t to be sampled
115  *
116  * This function allows callers to sample an errseq_t value, marking it as
117  * "seen" if required.
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 /**
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
142  *
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
145  * is no need to mark the value as seen.
146  *
147  * Returns the latest error set in the errseq_t or 0 if it hasn't changed.
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 /**
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
163  *
164  * Grab the eseq value, and see whether it matches the value that "since"
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.
175  */
176 int errseq_check_and_advance(errseq_t *eseq, errseq_t *since)
177 {
178 	int err = 0;
179 	errseq_t old, new;
180 
181 	/*
182 	 * Most callers will want to use the inline wrapper to check this,
183 	 * so that the common case of no error is handled without needing
184 	 * to take the lock that protects the "since" value.
185 	 */
186 	old = READ_ONCE(*eseq);
187 	if (old != *since) {
188 		/*
189 		 * Set the flag and try to swap it into place if it has
190 		 * changed.
191 		 *
192 		 * We don't care about the outcome of the swap here. If the
193 		 * swap doesn't occur, then it has either been updated by a
194 		 * writer who is altering the value in some way (updating
195 		 * counter or resetting the error), or another reader who is
196 		 * just setting the "seen" flag. Either outcome is OK, and we
197 		 * can advance "since" and return an error based on what we
198 		 * have.
199 		 */
200 		new = old | ERRSEQ_SEEN;
201 		if (new != old)
202 			cmpxchg(eseq, old, new);
203 		*since = new;
204 		err = -(new & MAX_ERRNO);
205 	}
206 	return err;
207 }
208 EXPORT_SYMBOL(errseq_check_and_advance);
209