3 #include <linux/kernel.h>
10 * RAS Correctable Errors Collector
12 * This is a simple gadget which collects correctable errors and counts their
13 * occurrence per physical page address.
15 * We've opted for possibly the simplest data structure to collect those - an
16 * array of the size of a memory page. It stores 512 u64's with the following
19 * [63 ... PFN ... 12 | 11 ... generation ... 10 | 9 ... count ... 0]
21 * The generation in the two highest order bits is two bits which are set to 11b
22 * on every insertion. During the course of each entry's existence, the
23 * generation field gets decremented during spring cleaning to 10b, then 01b and
26 * This way we're employing the natural numeric ordering to make sure that newly
27 * inserted/touched elements have higher 12-bit counts (which we've manufactured)
28 * and thus iterating over the array initially won't kick out those elements
29 * which were inserted last.
31 * Spring cleaning is what we do when we reach a certain number CLEAN_ELEMS of
32 * elements entered into the array, during which, we're decaying all elements.
33 * If, after decay, an element gets inserted again, its generation is set to 11b
34 * to make sure it has higher numerical count than other, older elements and
35 * thus emulate an an LRU-like behavior when deleting elements to free up space
38 * When an element reaches it's max count of count_threshold, we try to poison
39 * it by assuming that errors triggered count_threshold times in a single page
40 * are excessive and that page shouldn't be used anymore. count_threshold is
41 * initialized to COUNT_MASK which is the maximum.
43 * That error event entry causes cec_add_elem() to return !0 value and thus
44 * signal to its callers to log the error.
46 * To the question why we've chosen a page and moving elements around with
47 * memmove(), it is because it is a very simple structure to handle and max data
48 * movement is 4K which on highly optimized modern CPUs is almost unnoticeable.
49 * We wanted to avoid the pointer traversal of more complex structures like a
50 * linked list or some sort of a balancing search tree.
52 * Deleting an element takes O(n) but since it is only a single page, it should
53 * be fast enough and it shouldn't happen all too often depending on error
58 #define pr_fmt(fmt) "RAS: " fmt
61 * We use DECAY_BITS bits of PAGE_SHIFT bits for counting decay, i.e., how long
62 * elements have stayed in the array without having been accessed again.
65 #define DECAY_MASK ((1ULL << DECAY_BITS) - 1)
66 #define MAX_ELEMS (PAGE_SIZE / sizeof(u64))
69 * Threshold amount of inserted elements after which we start spring
72 #define CLEAN_ELEMS (MAX_ELEMS >> DECAY_BITS)
74 /* Bits which count the number of errors happened in this 4K page. */
75 #define COUNT_BITS (PAGE_SHIFT - DECAY_BITS)
76 #define COUNT_MASK ((1ULL << COUNT_BITS) - 1)
77 #define FULL_COUNT_MASK (PAGE_SIZE - 1)
80 * u64: [ 63 ... 12 | DECAY_BITS | COUNT_BITS ]
83 #define PFN(e) ((e) >> PAGE_SHIFT)
84 #define DECAY(e) (((e) >> COUNT_BITS) & DECAY_MASK)
85 #define COUNT(e) ((unsigned int)(e) & COUNT_MASK)
86 #define FULL_COUNT(e) ((e) & (PAGE_SIZE - 1))
88 static struct ce_array {
89 u64 *array; /* container page */
90 unsigned int n; /* number of elements in the array */
92 unsigned int decay_count; /*
93 * number of element insertions/increments
94 * since the last spring cleaning.
98 * number of PFNs which got poisoned.
102 * The number of correctable errors
103 * entered into the collector.
107 * Times we did spring cleaning.
112 __u32 disabled : 1, /* cmdline disabled */
119 static DEFINE_MUTEX(ce_mutex);
122 /* Amount of errors after which we offline */
123 static unsigned int count_threshold = COUNT_MASK;
126 * The timer "decays" element count each timer_interval which is 24hrs by
130 #define CEC_TIMER_DEFAULT_INTERVAL 24 * 60 * 60 /* 24 hrs */
131 #define CEC_TIMER_MIN_INTERVAL 1 * 60 * 60 /* 1h */
132 #define CEC_TIMER_MAX_INTERVAL 30 * 24 * 60 * 60 /* one month */
133 static struct timer_list cec_timer;
134 static u64 timer_interval = CEC_TIMER_DEFAULT_INTERVAL;
137 * Decrement decay value. We're using DECAY_BITS bits to denote decay of an
138 * element in the array. On insertion and any access, it gets reset to max.
140 static void do_spring_cleaning(struct ce_array *ca)
144 for (i = 0; i < ca->n; i++) {
145 u8 decay = DECAY(ca->array[i]);
152 ca->array[i] &= ~(DECAY_MASK << COUNT_BITS);
153 ca->array[i] |= (decay << COUNT_BITS);
160 * @interval in seconds
162 static void cec_mod_timer(struct timer_list *t, unsigned long interval)
166 iv = interval * HZ + jiffies;
168 mod_timer(t, round_jiffies(iv));
171 static void cec_timer_fn(unsigned long data)
173 struct ce_array *ca = (struct ce_array *)data;
175 do_spring_cleaning(ca);
177 cec_mod_timer(&cec_timer, timer_interval);
181 * @to: index of the smallest element which is >= then @pfn.
183 * Return the index of the pfn if found, otherwise negative value.
185 static int __find_elem(struct ce_array *ca, u64 pfn, unsigned int *to)
188 int min = 0, max = ca->n;
191 int tmp = (max + min) >> 1;
193 this_pfn = PFN(ca->array[tmp]);
197 else if (this_pfn > pfn)
208 this_pfn = PFN(ca->array[min]);
216 static int find_elem(struct ce_array *ca, u64 pfn, unsigned int *to)
224 return __find_elem(ca, pfn, to);
227 static void del_elem(struct ce_array *ca, int idx)
229 /* Save us a function call when deleting the last element. */
230 if (ca->n - (idx + 1))
231 memmove((void *)&ca->array[idx],
232 (void *)&ca->array[idx + 1],
233 (ca->n - (idx + 1)) * sizeof(u64));
238 static u64 del_lru_elem_unlocked(struct ce_array *ca)
240 unsigned int min = FULL_COUNT_MASK;
243 for (i = 0; i < ca->n; i++) {
244 unsigned int this = FULL_COUNT(ca->array[i]);
252 del_elem(ca, min_idx);
254 return PFN(ca->array[min_idx]);
258 * We return the 0th pfn in the error case under the assumption that it cannot
259 * be poisoned and excessive CEs in there are a serious deal anyway.
261 static u64 __maybe_unused del_lru_elem(void)
263 struct ce_array *ca = &ce_arr;
269 mutex_lock(&ce_mutex);
270 pfn = del_lru_elem_unlocked(ca);
271 mutex_unlock(&ce_mutex);
277 int cec_add_elem(u64 pfn)
279 struct ce_array *ca = &ce_arr;
284 * We can be called very early on the identify_cpu() path where we are
285 * not initialized yet. We ignore the error for simplicity.
287 if (!ce_arr.array || ce_arr.disabled)
292 mutex_lock(&ce_mutex);
294 if (ca->n == MAX_ELEMS)
295 WARN_ON(!del_lru_elem_unlocked(ca));
297 ret = find_elem(ca, pfn, &to);
300 * Shift range [to-end] to make room for one more element.
302 memmove((void *)&ca->array[to + 1],
303 (void *)&ca->array[to],
304 (ca->n - to) * sizeof(u64));
306 ca->array[to] = (pfn << PAGE_SHIFT) |
307 (DECAY_MASK << COUNT_BITS) | 1;
316 count = COUNT(ca->array[to]);
318 if (count < count_threshold) {
319 ca->array[to] |= (DECAY_MASK << COUNT_BITS);
324 u64 pfn = ca->array[to] >> PAGE_SHIFT;
326 if (!pfn_valid(pfn)) {
327 pr_warn("CEC: Invalid pfn: 0x%llx\n", pfn);
329 /* We have reached max count for this page, soft-offline it. */
330 pr_err("Soft-offlining pfn: 0x%llx\n", pfn);
331 memory_failure_queue(pfn, 0, MF_SOFT_OFFLINE);
338 * Return a >0 value to denote that we've reached the offlining
349 if (ca->decay_count >= CLEAN_ELEMS)
350 do_spring_cleaning(ca);
353 mutex_unlock(&ce_mutex);
358 static int u64_get(void *data, u64 *val)
365 static int pfn_set(void *data, u64 val)
369 return cec_add_elem(val);
372 DEFINE_DEBUGFS_ATTRIBUTE(pfn_ops, u64_get, pfn_set, "0x%llx\n");
374 static int decay_interval_set(void *data, u64 val)
378 if (val < CEC_TIMER_MIN_INTERVAL)
381 if (val > CEC_TIMER_MAX_INTERVAL)
384 timer_interval = val;
386 cec_mod_timer(&cec_timer, timer_interval);
389 DEFINE_DEBUGFS_ATTRIBUTE(decay_interval_ops, u64_get, decay_interval_set, "%lld\n");
391 static int count_threshold_set(void *data, u64 val)
395 if (val > COUNT_MASK)
398 count_threshold = val;
402 DEFINE_DEBUGFS_ATTRIBUTE(count_threshold_ops, u64_get, count_threshold_set, "%lld\n");
404 static int array_dump(struct seq_file *m, void *v)
406 struct ce_array *ca = &ce_arr;
410 mutex_lock(&ce_mutex);
412 seq_printf(m, "{ n: %d\n", ca->n);
413 for (i = 0; i < ca->n; i++) {
414 u64 this = PFN(ca->array[i]);
416 seq_printf(m, " %03d: [%016llx|%03llx]\n", i, this, FULL_COUNT(ca->array[i]));
418 WARN_ON(prev > this);
423 seq_printf(m, "}\n");
425 seq_printf(m, "Stats:\nCEs: %llu\nofflined pages: %llu\n",
426 ca->ces_entered, ca->pfns_poisoned);
428 seq_printf(m, "Flags: 0x%x\n", ca->flags);
430 seq_printf(m, "Timer interval: %lld seconds\n", timer_interval);
431 seq_printf(m, "Decays: %lld\n", ca->decays_done);
433 seq_printf(m, "Action threshold: %d\n", count_threshold);
435 mutex_unlock(&ce_mutex);
440 static int array_open(struct inode *inode, struct file *filp)
442 return single_open(filp, array_dump, NULL);
445 static const struct file_operations array_ops = {
446 .owner = THIS_MODULE,
450 .release = single_release,
453 static int __init create_debugfs_nodes(void)
455 struct dentry *d, *pfn, *decay, *count, *array;
457 d = debugfs_create_dir("cec", ras_debugfs_dir);
459 pr_warn("Error creating cec debugfs node!\n");
463 pfn = debugfs_create_file("pfn", S_IRUSR | S_IWUSR, d, &dfs_pfn, &pfn_ops);
465 pr_warn("Error creating pfn debugfs node!\n");
469 array = debugfs_create_file("array", S_IRUSR, d, NULL, &array_ops);
471 pr_warn("Error creating array debugfs node!\n");
475 decay = debugfs_create_file("decay_interval", S_IRUSR | S_IWUSR, d,
476 &timer_interval, &decay_interval_ops);
478 pr_warn("Error creating decay_interval debugfs node!\n");
482 count = debugfs_create_file("count_threshold", S_IRUSR | S_IWUSR, d,
483 &count_threshold, &count_threshold_ops);
485 pr_warn("Error creating count_threshold debugfs node!\n");
493 debugfs_remove_recursive(d);
498 void __init cec_init(void)
503 ce_arr.array = (void *)get_zeroed_page(GFP_KERNEL);
505 pr_err("Error allocating CE array page!\n");
509 if (create_debugfs_nodes())
512 setup_timer(&cec_timer, cec_timer_fn, (unsigned long)&ce_arr);
513 cec_mod_timer(&cec_timer, CEC_TIMER_DEFAULT_INTERVAL);
515 pr_info("Correctable Errors collector initialized.\n");
518 int __init parse_cec_param(char *str)
526 if (!strncmp(str, "cec_disable", 7))