1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2017-2019 Borislav Petkov, SUSE Labs.
7 #include <linux/kernel.h>
8 #include <linux/workqueue.h>
15 * RAS Correctable Errors Collector
17 * This is a simple gadget which collects correctable errors and counts their
18 * occurrence per physical page address.
20 * We've opted for possibly the simplest data structure to collect those - an
21 * array of the size of a memory page. It stores 512 u64's with the following
24 * [63 ... PFN ... 12 | 11 ... generation ... 10 | 9 ... count ... 0]
26 * The generation in the two highest order bits is two bits which are set to 11b
27 * on every insertion. During the course of each entry's existence, the
28 * generation field gets decremented during spring cleaning to 10b, then 01b and
31 * This way we're employing the natural numeric ordering to make sure that newly
32 * inserted/touched elements have higher 12-bit counts (which we've manufactured)
33 * and thus iterating over the array initially won't kick out those elements
34 * which were inserted last.
36 * Spring cleaning is what we do when we reach a certain number CLEAN_ELEMS of
37 * elements entered into the array, during which, we're decaying all elements.
38 * If, after decay, an element gets inserted again, its generation is set to 11b
39 * to make sure it has higher numerical count than other, older elements and
40 * thus emulate an an LRU-like behavior when deleting elements to free up space
43 * When an element reaches it's max count of action_threshold, we try to poison
44 * it by assuming that errors triggered action_threshold times in a single page
45 * are excessive and that page shouldn't be used anymore. action_threshold is
46 * initialized to COUNT_MASK which is the maximum.
48 * That error event entry causes cec_add_elem() to return !0 value and thus
49 * signal to its callers to log the error.
51 * To the question why we've chosen a page and moving elements around with
52 * memmove(), it is because it is a very simple structure to handle and max data
53 * movement is 4K which on highly optimized modern CPUs is almost unnoticeable.
54 * We wanted to avoid the pointer traversal of more complex structures like a
55 * linked list or some sort of a balancing search tree.
57 * Deleting an element takes O(n) but since it is only a single page, it should
58 * be fast enough and it shouldn't happen all too often depending on error
63 #define pr_fmt(fmt) "RAS: " fmt
66 * We use DECAY_BITS bits of PAGE_SHIFT bits for counting decay, i.e., how long
67 * elements have stayed in the array without having been accessed again.
70 #define DECAY_MASK ((1ULL << DECAY_BITS) - 1)
71 #define MAX_ELEMS (PAGE_SIZE / sizeof(u64))
74 * Threshold amount of inserted elements after which we start spring
77 #define CLEAN_ELEMS (MAX_ELEMS >> DECAY_BITS)
79 /* Bits which count the number of errors happened in this 4K page. */
80 #define COUNT_BITS (PAGE_SHIFT - DECAY_BITS)
81 #define COUNT_MASK ((1ULL << COUNT_BITS) - 1)
82 #define FULL_COUNT_MASK (PAGE_SIZE - 1)
85 * u64: [ 63 ... 12 | DECAY_BITS | COUNT_BITS ]
88 #define PFN(e) ((e) >> PAGE_SHIFT)
89 #define DECAY(e) (((e) >> COUNT_BITS) & DECAY_MASK)
90 #define COUNT(e) ((unsigned int)(e) & COUNT_MASK)
91 #define FULL_COUNT(e) ((e) & (PAGE_SIZE - 1))
93 static struct ce_array {
94 u64 *array; /* container page */
95 unsigned int n; /* number of elements in the array */
97 unsigned int decay_count; /*
98 * number of element insertions/increments
99 * since the last spring cleaning.
102 u64 pfns_poisoned; /*
103 * number of PFNs which got poisoned.
107 * The number of correctable errors
108 * entered into the collector.
112 * Times we did spring cleaning.
117 __u32 disabled : 1, /* cmdline disabled */
124 static DEFINE_MUTEX(ce_mutex);
127 /* Amount of errors after which we offline */
128 static u64 action_threshold = COUNT_MASK;
130 /* Each element "decays" each decay_interval which is 24hrs by default. */
131 #define CEC_DECAY_DEFAULT_INTERVAL 24 * 60 * 60 /* 24 hrs */
132 #define CEC_DECAY_MIN_INTERVAL 1 * 60 * 60 /* 1h */
133 #define CEC_DECAY_MAX_INTERVAL 30 * 24 * 60 * 60 /* one month */
134 static struct delayed_work cec_work;
135 static u64 decay_interval = CEC_DECAY_DEFAULT_INTERVAL;
138 * Decrement decay value. We're using DECAY_BITS bits to denote decay of an
139 * element in the array. On insertion and any access, it gets reset to max.
141 static void do_spring_cleaning(struct ce_array *ca)
145 for (i = 0; i < ca->n; i++) {
146 u8 decay = DECAY(ca->array[i]);
153 ca->array[i] &= ~(DECAY_MASK << COUNT_BITS);
154 ca->array[i] |= (decay << COUNT_BITS);
161 * @interval in seconds
163 static void cec_mod_work(unsigned long interval)
168 mod_delayed_work(system_wq, &cec_work, round_jiffies(iv));
171 static void cec_work_fn(struct work_struct *work)
173 mutex_lock(&ce_mutex);
174 do_spring_cleaning(&ce_arr);
175 mutex_unlock(&ce_mutex);
177 cec_mod_work(decay_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)
187 int min = 0, max = ca->n - 1;
191 int i = (min + max) >> 1;
193 this_pfn = PFN(ca->array[i]);
197 else if (this_pfn > pfn)
199 else if (this_pfn == pfn) {
208 * When the loop terminates without finding @pfn, min has the index of
209 * the element slot where the new @pfn should be inserted. The loop
210 * terminates when min > max, which means the min index points to the
211 * bigger element while the max index to the smaller element, in-between
212 * which the new @pfn belongs to.
214 * For more details, see exercise 1, Section 6.2.1 in TAOCP, vol. 3.
222 static int find_elem(struct ce_array *ca, u64 pfn, unsigned int *to)
230 return __find_elem(ca, pfn, to);
233 static void del_elem(struct ce_array *ca, int idx)
235 /* Save us a function call when deleting the last element. */
236 if (ca->n - (idx + 1))
237 memmove((void *)&ca->array[idx],
238 (void *)&ca->array[idx + 1],
239 (ca->n - (idx + 1)) * sizeof(u64));
244 static u64 del_lru_elem_unlocked(struct ce_array *ca)
246 unsigned int min = FULL_COUNT_MASK;
249 for (i = 0; i < ca->n; i++) {
250 unsigned int this = FULL_COUNT(ca->array[i]);
258 del_elem(ca, min_idx);
260 return PFN(ca->array[min_idx]);
264 * We return the 0th pfn in the error case under the assumption that it cannot
265 * be poisoned and excessive CEs in there are a serious deal anyway.
267 static u64 __maybe_unused del_lru_elem(void)
269 struct ce_array *ca = &ce_arr;
275 mutex_lock(&ce_mutex);
276 pfn = del_lru_elem_unlocked(ca);
277 mutex_unlock(&ce_mutex);
282 static bool sanity_check(struct ce_array *ca)
288 for (i = 0; i < ca->n; i++) {
289 u64 this = PFN(ca->array[i]);
291 if (WARN(prev > this, "prev: 0x%016llx <-> this: 0x%016llx\n", prev, this))
300 pr_info("Sanity check dump:\n{ n: %d\n", ca->n);
301 for (i = 0; i < ca->n; i++) {
302 u64 this = PFN(ca->array[i]);
304 pr_info(" %03d: [%016llx|%03llx]\n", i, this, FULL_COUNT(ca->array[i]));
311 int cec_add_elem(u64 pfn)
313 struct ce_array *ca = &ce_arr;
318 * We can be called very early on the identify_cpu() path where we are
319 * not initialized yet. We ignore the error for simplicity.
321 if (!ce_arr.array || ce_arr.disabled)
324 mutex_lock(&ce_mutex);
328 /* Array full, free the LRU slot. */
329 if (ca->n == MAX_ELEMS)
330 WARN_ON(!del_lru_elem_unlocked(ca));
332 ret = find_elem(ca, pfn, &to);
335 * Shift range [to-end] to make room for one more element.
337 memmove((void *)&ca->array[to + 1],
338 (void *)&ca->array[to],
339 (ca->n - to) * sizeof(u64));
341 ca->array[to] = pfn << PAGE_SHIFT;
345 /* Add/refresh element generation and increment count */
346 ca->array[to] |= DECAY_MASK << COUNT_BITS;
349 /* Check action threshold and soft-offline, if reached. */
350 count = COUNT(ca->array[to]);
351 if (count >= action_threshold) {
352 u64 pfn = ca->array[to] >> PAGE_SHIFT;
354 if (!pfn_valid(pfn)) {
355 pr_warn("CEC: Invalid pfn: 0x%llx\n", pfn);
357 /* We have reached max count for this page, soft-offline it. */
358 pr_err("Soft-offlining pfn: 0x%llx\n", pfn);
359 memory_failure_queue(pfn, MF_SOFT_OFFLINE);
366 * Return a >0 value to callers, to denote that we've reached
367 * the offlining threshold.
376 if (ca->decay_count >= CLEAN_ELEMS)
377 do_spring_cleaning(ca);
379 WARN_ON_ONCE(sanity_check(ca));
382 mutex_unlock(&ce_mutex);
387 static int u64_get(void *data, u64 *val)
394 static int pfn_set(void *data, u64 val)
403 DEFINE_DEBUGFS_ATTRIBUTE(pfn_ops, u64_get, pfn_set, "0x%llx\n");
405 static int decay_interval_set(void *data, u64 val)
407 if (val < CEC_DECAY_MIN_INTERVAL)
410 if (val > CEC_DECAY_MAX_INTERVAL)
414 decay_interval = val;
416 cec_mod_work(decay_interval);
420 DEFINE_DEBUGFS_ATTRIBUTE(decay_interval_ops, u64_get, decay_interval_set, "%lld\n");
422 static int action_threshold_set(void *data, u64 val)
426 if (val > COUNT_MASK)
429 action_threshold = val;
433 DEFINE_DEBUGFS_ATTRIBUTE(action_threshold_ops, u64_get, action_threshold_set, "%lld\n");
435 static const char * const bins[] = { "00", "01", "10", "11" };
437 static int array_dump(struct seq_file *m, void *v)
439 struct ce_array *ca = &ce_arr;
442 mutex_lock(&ce_mutex);
444 seq_printf(m, "{ n: %d\n", ca->n);
445 for (i = 0; i < ca->n; i++) {
446 u64 this = PFN(ca->array[i]);
448 seq_printf(m, " %3d: [%016llx|%s|%03llx]\n",
449 i, this, bins[DECAY(ca->array[i])], COUNT(ca->array[i]));
452 seq_printf(m, "}\n");
454 seq_printf(m, "Stats:\nCEs: %llu\nofflined pages: %llu\n",
455 ca->ces_entered, ca->pfns_poisoned);
457 seq_printf(m, "Flags: 0x%x\n", ca->flags);
459 seq_printf(m, "Decay interval: %lld seconds\n", decay_interval);
460 seq_printf(m, "Decays: %lld\n", ca->decays_done);
462 seq_printf(m, "Action threshold: %lld\n", action_threshold);
464 mutex_unlock(&ce_mutex);
469 static int array_open(struct inode *inode, struct file *filp)
471 return single_open(filp, array_dump, NULL);
474 static const struct file_operations array_ops = {
475 .owner = THIS_MODULE,
479 .release = single_release,
482 static int __init create_debugfs_nodes(void)
484 struct dentry *d, *pfn, *decay, *count, *array;
486 d = debugfs_create_dir("cec", ras_debugfs_dir);
488 pr_warn("Error creating cec debugfs node!\n");
492 decay = debugfs_create_file("decay_interval", S_IRUSR | S_IWUSR, d,
493 &decay_interval, &decay_interval_ops);
495 pr_warn("Error creating decay_interval debugfs node!\n");
499 count = debugfs_create_file("action_threshold", S_IRUSR | S_IWUSR, d,
500 &action_threshold, &action_threshold_ops);
502 pr_warn("Error creating action_threshold debugfs node!\n");
506 if (!IS_ENABLED(CONFIG_RAS_CEC_DEBUG))
509 pfn = debugfs_create_file("pfn", S_IRUSR | S_IWUSR, d, &dfs_pfn, &pfn_ops);
511 pr_warn("Error creating pfn debugfs node!\n");
515 array = debugfs_create_file("array", S_IRUSR, d, NULL, &array_ops);
517 pr_warn("Error creating array debugfs node!\n");
524 debugfs_remove_recursive(d);
529 void __init cec_init(void)
534 ce_arr.array = (void *)get_zeroed_page(GFP_KERNEL);
536 pr_err("Error allocating CE array page!\n");
540 if (create_debugfs_nodes()) {
541 free_page((unsigned long)ce_arr.array);
545 INIT_DELAYED_WORK(&cec_work, cec_work_fn);
546 schedule_delayed_work(&cec_work, CEC_DECAY_DEFAULT_INTERVAL);
548 pr_info("Correctable Errors collector initialized.\n");
551 int __init parse_cec_param(char *str)
559 if (!strcmp(str, "cec_disable"))