]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
011d8261 BP |
2 | #include <linux/mm.h> |
3 | #include <linux/gfp.h> | |
4 | #include <linux/kernel.h> | |
2328874d | 5 | #include <linux/workqueue.h> |
011d8261 BP |
6 | |
7 | #include <asm/mce.h> | |
8 | ||
9 | #include "debugfs.h" | |
10 | ||
11 | /* | |
12 | * RAS Correctable Errors Collector | |
13 | * | |
14 | * This is a simple gadget which collects correctable errors and counts their | |
15 | * occurrence per physical page address. | |
16 | * | |
17 | * We've opted for possibly the simplest data structure to collect those - an | |
18 | * array of the size of a memory page. It stores 512 u64's with the following | |
19 | * structure: | |
20 | * | |
21 | * [63 ... PFN ... 12 | 11 ... generation ... 10 | 9 ... count ... 0] | |
22 | * | |
23 | * The generation in the two highest order bits is two bits which are set to 11b | |
24 | * on every insertion. During the course of each entry's existence, the | |
25 | * generation field gets decremented during spring cleaning to 10b, then 01b and | |
26 | * then 00b. | |
27 | * | |
28 | * This way we're employing the natural numeric ordering to make sure that newly | |
29 | * inserted/touched elements have higher 12-bit counts (which we've manufactured) | |
30 | * and thus iterating over the array initially won't kick out those elements | |
31 | * which were inserted last. | |
32 | * | |
33 | * Spring cleaning is what we do when we reach a certain number CLEAN_ELEMS of | |
34 | * elements entered into the array, during which, we're decaying all elements. | |
35 | * If, after decay, an element gets inserted again, its generation is set to 11b | |
36 | * to make sure it has higher numerical count than other, older elements and | |
37 | * thus emulate an an LRU-like behavior when deleting elements to free up space | |
38 | * in the page. | |
39 | * | |
40 | * When an element reaches it's max count of count_threshold, we try to poison | |
41 | * it by assuming that errors triggered count_threshold times in a single page | |
42 | * are excessive and that page shouldn't be used anymore. count_threshold is | |
43 | * initialized to COUNT_MASK which is the maximum. | |
44 | * | |
45 | * That error event entry causes cec_add_elem() to return !0 value and thus | |
46 | * signal to its callers to log the error. | |
47 | * | |
48 | * To the question why we've chosen a page and moving elements around with | |
49 | * memmove(), it is because it is a very simple structure to handle and max data | |
50 | * movement is 4K which on highly optimized modern CPUs is almost unnoticeable. | |
51 | * We wanted to avoid the pointer traversal of more complex structures like a | |
52 | * linked list or some sort of a balancing search tree. | |
53 | * | |
54 | * Deleting an element takes O(n) but since it is only a single page, it should | |
55 | * be fast enough and it shouldn't happen all too often depending on error | |
56 | * patterns. | |
57 | */ | |
58 | ||
59 | #undef pr_fmt | |
60 | #define pr_fmt(fmt) "RAS: " fmt | |
61 | ||
62 | /* | |
63 | * We use DECAY_BITS bits of PAGE_SHIFT bits for counting decay, i.e., how long | |
64 | * elements have stayed in the array without having been accessed again. | |
65 | */ | |
66 | #define DECAY_BITS 2 | |
67 | #define DECAY_MASK ((1ULL << DECAY_BITS) - 1) | |
68 | #define MAX_ELEMS (PAGE_SIZE / sizeof(u64)) | |
69 | ||
70 | /* | |
71 | * Threshold amount of inserted elements after which we start spring | |
72 | * cleaning. | |
73 | */ | |
74 | #define CLEAN_ELEMS (MAX_ELEMS >> DECAY_BITS) | |
75 | ||
76 | /* Bits which count the number of errors happened in this 4K page. */ | |
77 | #define COUNT_BITS (PAGE_SHIFT - DECAY_BITS) | |
78 | #define COUNT_MASK ((1ULL << COUNT_BITS) - 1) | |
79 | #define FULL_COUNT_MASK (PAGE_SIZE - 1) | |
80 | ||
81 | /* | |
82 | * u64: [ 63 ... 12 | DECAY_BITS | COUNT_BITS ] | |
83 | */ | |
84 | ||
85 | #define PFN(e) ((e) >> PAGE_SHIFT) | |
86 | #define DECAY(e) (((e) >> COUNT_BITS) & DECAY_MASK) | |
87 | #define COUNT(e) ((unsigned int)(e) & COUNT_MASK) | |
88 | #define FULL_COUNT(e) ((e) & (PAGE_SIZE - 1)) | |
89 | ||
90 | static struct ce_array { | |
91 | u64 *array; /* container page */ | |
92 | unsigned int n; /* number of elements in the array */ | |
93 | ||
94 | unsigned int decay_count; /* | |
95 | * number of element insertions/increments | |
96 | * since the last spring cleaning. | |
97 | */ | |
98 | ||
99 | u64 pfns_poisoned; /* | |
100 | * number of PFNs which got poisoned. | |
101 | */ | |
102 | ||
103 | u64 ces_entered; /* | |
104 | * The number of correctable errors | |
105 | * entered into the collector. | |
106 | */ | |
107 | ||
108 | u64 decays_done; /* | |
109 | * Times we did spring cleaning. | |
110 | */ | |
111 | ||
112 | union { | |
113 | struct { | |
114 | __u32 disabled : 1, /* cmdline disabled */ | |
115 | __resv : 31; | |
116 | }; | |
117 | __u32 flags; | |
118 | }; | |
119 | } ce_arr; | |
120 | ||
121 | static DEFINE_MUTEX(ce_mutex); | |
122 | static u64 dfs_pfn; | |
123 | ||
124 | /* Amount of errors after which we offline */ | |
125 | static unsigned int count_threshold = COUNT_MASK; | |
126 | ||
2328874d CW |
127 | /* Each element "decays" each decay_interval which is 24hrs by default. */ |
128 | #define CEC_DECAY_DEFAULT_INTERVAL 24 * 60 * 60 /* 24 hrs */ | |
129 | #define CEC_DECAY_MIN_INTERVAL 1 * 60 * 60 /* 1h */ | |
130 | #define CEC_DECAY_MAX_INTERVAL 30 * 24 * 60 * 60 /* one month */ | |
131 | static struct delayed_work cec_work; | |
132 | static u64 decay_interval = CEC_DECAY_DEFAULT_INTERVAL; | |
011d8261 BP |
133 | |
134 | /* | |
135 | * Decrement decay value. We're using DECAY_BITS bits to denote decay of an | |
136 | * element in the array. On insertion and any access, it gets reset to max. | |
137 | */ | |
138 | static void do_spring_cleaning(struct ce_array *ca) | |
139 | { | |
140 | int i; | |
141 | ||
142 | for (i = 0; i < ca->n; i++) { | |
143 | u8 decay = DECAY(ca->array[i]); | |
144 | ||
145 | if (!decay) | |
146 | continue; | |
147 | ||
148 | decay--; | |
149 | ||
150 | ca->array[i] &= ~(DECAY_MASK << COUNT_BITS); | |
151 | ca->array[i] |= (decay << COUNT_BITS); | |
152 | } | |
153 | ca->decay_count = 0; | |
154 | ca->decays_done++; | |
155 | } | |
156 | ||
157 | /* | |
158 | * @interval in seconds | |
159 | */ | |
2328874d | 160 | static void cec_mod_work(unsigned long interval) |
011d8261 BP |
161 | { |
162 | unsigned long iv; | |
163 | ||
2328874d CW |
164 | iv = interval * HZ; |
165 | mod_delayed_work(system_wq, &cec_work, round_jiffies(iv)); | |
011d8261 BP |
166 | } |
167 | ||
2328874d | 168 | static void cec_work_fn(struct work_struct *work) |
011d8261 | 169 | { |
2328874d | 170 | mutex_lock(&ce_mutex); |
254db5bd | 171 | do_spring_cleaning(&ce_arr); |
2328874d | 172 | mutex_unlock(&ce_mutex); |
011d8261 | 173 | |
2328874d | 174 | cec_mod_work(decay_interval); |
011d8261 BP |
175 | } |
176 | ||
177 | /* | |
178 | * @to: index of the smallest element which is >= then @pfn. | |
179 | * | |
180 | * Return the index of the pfn if found, otherwise negative value. | |
181 | */ | |
182 | static int __find_elem(struct ce_array *ca, u64 pfn, unsigned int *to) | |
183 | { | |
61d9d59b | 184 | int min = 0, max = ca->n - 1; |
011d8261 | 185 | u64 this_pfn; |
011d8261 | 186 | |
61d9d59b BP |
187 | while (min <= max) { |
188 | int i = (min + max) >> 1; | |
011d8261 | 189 | |
61d9d59b | 190 | this_pfn = PFN(ca->array[i]); |
011d8261 BP |
191 | |
192 | if (this_pfn < pfn) | |
61d9d59b | 193 | min = i + 1; |
011d8261 | 194 | else if (this_pfn > pfn) |
61d9d59b BP |
195 | max = i - 1; |
196 | else if (this_pfn == pfn) { | |
197 | if (to) | |
198 | *to = i; | |
199 | ||
200 | return i; | |
011d8261 BP |
201 | } |
202 | } | |
203 | ||
61d9d59b BP |
204 | /* |
205 | * When the loop terminates without finding @pfn, min has the index of | |
206 | * the element slot where the new @pfn should be inserted. The loop | |
207 | * terminates when min > max, which means the min index points to the | |
208 | * bigger element while the max index to the smaller element, in-between | |
209 | * which the new @pfn belongs to. | |
210 | * | |
211 | * For more details, see exercise 1, Section 6.2.1 in TAOCP, vol. 3. | |
212 | */ | |
011d8261 BP |
213 | if (to) |
214 | *to = min; | |
215 | ||
011d8261 BP |
216 | return -ENOKEY; |
217 | } | |
218 | ||
219 | static int find_elem(struct ce_array *ca, u64 pfn, unsigned int *to) | |
220 | { | |
221 | WARN_ON(!to); | |
222 | ||
223 | if (!ca->n) { | |
224 | *to = 0; | |
225 | return -ENOKEY; | |
226 | } | |
227 | return __find_elem(ca, pfn, to); | |
228 | } | |
229 | ||
230 | static void del_elem(struct ce_array *ca, int idx) | |
231 | { | |
232 | /* Save us a function call when deleting the last element. */ | |
233 | if (ca->n - (idx + 1)) | |
234 | memmove((void *)&ca->array[idx], | |
235 | (void *)&ca->array[idx + 1], | |
236 | (ca->n - (idx + 1)) * sizeof(u64)); | |
237 | ||
238 | ca->n--; | |
239 | } | |
240 | ||
241 | static u64 del_lru_elem_unlocked(struct ce_array *ca) | |
242 | { | |
243 | unsigned int min = FULL_COUNT_MASK; | |
244 | int i, min_idx = 0; | |
245 | ||
246 | for (i = 0; i < ca->n; i++) { | |
247 | unsigned int this = FULL_COUNT(ca->array[i]); | |
248 | ||
249 | if (min > this) { | |
250 | min = this; | |
251 | min_idx = i; | |
252 | } | |
253 | } | |
254 | ||
255 | del_elem(ca, min_idx); | |
256 | ||
257 | return PFN(ca->array[min_idx]); | |
258 | } | |
259 | ||
260 | /* | |
261 | * We return the 0th pfn in the error case under the assumption that it cannot | |
262 | * be poisoned and excessive CEs in there are a serious deal anyway. | |
263 | */ | |
264 | static u64 __maybe_unused del_lru_elem(void) | |
265 | { | |
266 | struct ce_array *ca = &ce_arr; | |
267 | u64 pfn; | |
268 | ||
269 | if (!ca->n) | |
270 | return 0; | |
271 | ||
272 | mutex_lock(&ce_mutex); | |
273 | pfn = del_lru_elem_unlocked(ca); | |
274 | mutex_unlock(&ce_mutex); | |
275 | ||
276 | return pfn; | |
277 | } | |
278 | ||
279 | ||
280 | int cec_add_elem(u64 pfn) | |
281 | { | |
282 | struct ce_array *ca = &ce_arr; | |
283 | unsigned int to; | |
284 | int count, ret = 0; | |
285 | ||
286 | /* | |
287 | * We can be called very early on the identify_cpu() path where we are | |
288 | * not initialized yet. We ignore the error for simplicity. | |
289 | */ | |
290 | if (!ce_arr.array || ce_arr.disabled) | |
291 | return -ENODEV; | |
292 | ||
293 | ca->ces_entered++; | |
294 | ||
295 | mutex_lock(&ce_mutex); | |
296 | ||
297 | if (ca->n == MAX_ELEMS) | |
298 | WARN_ON(!del_lru_elem_unlocked(ca)); | |
299 | ||
300 | ret = find_elem(ca, pfn, &to); | |
301 | if (ret < 0) { | |
302 | /* | |
303 | * Shift range [to-end] to make room for one more element. | |
304 | */ | |
305 | memmove((void *)&ca->array[to + 1], | |
306 | (void *)&ca->array[to], | |
307 | (ca->n - to) * sizeof(u64)); | |
308 | ||
309 | ca->array[to] = (pfn << PAGE_SHIFT) | | |
310 | (DECAY_MASK << COUNT_BITS) | 1; | |
311 | ||
312 | ca->n++; | |
313 | ||
314 | ret = 0; | |
315 | ||
316 | goto decay; | |
317 | } | |
318 | ||
319 | count = COUNT(ca->array[to]); | |
320 | ||
321 | if (count < count_threshold) { | |
322 | ca->array[to] |= (DECAY_MASK << COUNT_BITS); | |
323 | ca->array[to]++; | |
324 | ||
325 | ret = 0; | |
326 | } else { | |
327 | u64 pfn = ca->array[to] >> PAGE_SHIFT; | |
328 | ||
329 | if (!pfn_valid(pfn)) { | |
330 | pr_warn("CEC: Invalid pfn: 0x%llx\n", pfn); | |
331 | } else { | |
332 | /* We have reached max count for this page, soft-offline it. */ | |
333 | pr_err("Soft-offlining pfn: 0x%llx\n", pfn); | |
334 | memory_failure_queue(pfn, 0, MF_SOFT_OFFLINE); | |
335 | ca->pfns_poisoned++; | |
336 | } | |
337 | ||
338 | del_elem(ca, to); | |
339 | ||
340 | /* | |
341 | * Return a >0 value to denote that we've reached the offlining | |
342 | * threshold. | |
343 | */ | |
344 | ret = 1; | |
345 | ||
346 | goto unlock; | |
347 | } | |
348 | ||
349 | decay: | |
350 | ca->decay_count++; | |
351 | ||
352 | if (ca->decay_count >= CLEAN_ELEMS) | |
353 | do_spring_cleaning(ca); | |
354 | ||
355 | unlock: | |
356 | mutex_unlock(&ce_mutex); | |
357 | ||
358 | return ret; | |
359 | } | |
360 | ||
361 | static int u64_get(void *data, u64 *val) | |
362 | { | |
363 | *val = *(u64 *)data; | |
364 | ||
365 | return 0; | |
366 | } | |
367 | ||
368 | static int pfn_set(void *data, u64 val) | |
369 | { | |
370 | *(u64 *)data = val; | |
371 | ||
ca5195c7 BP |
372 | cec_add_elem(val); |
373 | ||
374 | return 0; | |
011d8261 BP |
375 | } |
376 | ||
377 | DEFINE_DEBUGFS_ATTRIBUTE(pfn_ops, u64_get, pfn_set, "0x%llx\n"); | |
378 | ||
379 | static int decay_interval_set(void *data, u64 val) | |
380 | { | |
381 | *(u64 *)data = val; | |
382 | ||
2328874d | 383 | if (val < CEC_DECAY_MIN_INTERVAL) |
011d8261 BP |
384 | return -EINVAL; |
385 | ||
2328874d | 386 | if (val > CEC_DECAY_MAX_INTERVAL) |
011d8261 BP |
387 | return -EINVAL; |
388 | ||
2328874d | 389 | decay_interval = val; |
011d8261 | 390 | |
2328874d | 391 | cec_mod_work(decay_interval); |
011d8261 BP |
392 | return 0; |
393 | } | |
394 | DEFINE_DEBUGFS_ATTRIBUTE(decay_interval_ops, u64_get, decay_interval_set, "%lld\n"); | |
395 | ||
396 | static int count_threshold_set(void *data, u64 val) | |
397 | { | |
398 | *(u64 *)data = val; | |
399 | ||
400 | if (val > COUNT_MASK) | |
401 | val = COUNT_MASK; | |
402 | ||
403 | count_threshold = val; | |
404 | ||
405 | return 0; | |
406 | } | |
407 | DEFINE_DEBUGFS_ATTRIBUTE(count_threshold_ops, u64_get, count_threshold_set, "%lld\n"); | |
408 | ||
409 | static int array_dump(struct seq_file *m, void *v) | |
410 | { | |
411 | struct ce_array *ca = &ce_arr; | |
412 | u64 prev = 0; | |
413 | int i; | |
414 | ||
415 | mutex_lock(&ce_mutex); | |
416 | ||
417 | seq_printf(m, "{ n: %d\n", ca->n); | |
418 | for (i = 0; i < ca->n; i++) { | |
419 | u64 this = PFN(ca->array[i]); | |
420 | ||
421 | seq_printf(m, " %03d: [%016llx|%03llx]\n", i, this, FULL_COUNT(ca->array[i])); | |
422 | ||
423 | WARN_ON(prev > this); | |
424 | ||
425 | prev = this; | |
426 | } | |
427 | ||
428 | seq_printf(m, "}\n"); | |
429 | ||
430 | seq_printf(m, "Stats:\nCEs: %llu\nofflined pages: %llu\n", | |
431 | ca->ces_entered, ca->pfns_poisoned); | |
432 | ||
433 | seq_printf(m, "Flags: 0x%x\n", ca->flags); | |
434 | ||
2328874d | 435 | seq_printf(m, "Decay interval: %lld seconds\n", decay_interval); |
011d8261 BP |
436 | seq_printf(m, "Decays: %lld\n", ca->decays_done); |
437 | ||
438 | seq_printf(m, "Action threshold: %d\n", count_threshold); | |
439 | ||
440 | mutex_unlock(&ce_mutex); | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
445 | static int array_open(struct inode *inode, struct file *filp) | |
446 | { | |
447 | return single_open(filp, array_dump, NULL); | |
448 | } | |
449 | ||
450 | static const struct file_operations array_ops = { | |
451 | .owner = THIS_MODULE, | |
452 | .open = array_open, | |
453 | .read = seq_read, | |
454 | .llseek = seq_lseek, | |
455 | .release = single_release, | |
456 | }; | |
457 | ||
458 | static int __init create_debugfs_nodes(void) | |
459 | { | |
460 | struct dentry *d, *pfn, *decay, *count, *array; | |
461 | ||
462 | d = debugfs_create_dir("cec", ras_debugfs_dir); | |
463 | if (!d) { | |
464 | pr_warn("Error creating cec debugfs node!\n"); | |
465 | return -1; | |
466 | } | |
467 | ||
468 | pfn = debugfs_create_file("pfn", S_IRUSR | S_IWUSR, d, &dfs_pfn, &pfn_ops); | |
469 | if (!pfn) { | |
470 | pr_warn("Error creating pfn debugfs node!\n"); | |
471 | goto err; | |
472 | } | |
473 | ||
474 | array = debugfs_create_file("array", S_IRUSR, d, NULL, &array_ops); | |
475 | if (!array) { | |
476 | pr_warn("Error creating array debugfs node!\n"); | |
477 | goto err; | |
478 | } | |
479 | ||
480 | decay = debugfs_create_file("decay_interval", S_IRUSR | S_IWUSR, d, | |
2328874d | 481 | &decay_interval, &decay_interval_ops); |
011d8261 BP |
482 | if (!decay) { |
483 | pr_warn("Error creating decay_interval debugfs node!\n"); | |
484 | goto err; | |
485 | } | |
486 | ||
487 | count = debugfs_create_file("count_threshold", S_IRUSR | S_IWUSR, d, | |
488 | &count_threshold, &count_threshold_ops); | |
32288daf | 489 | if (!count) { |
011d8261 BP |
490 | pr_warn("Error creating count_threshold debugfs node!\n"); |
491 | goto err; | |
492 | } | |
493 | ||
494 | ||
495 | return 0; | |
496 | ||
497 | err: | |
498 | debugfs_remove_recursive(d); | |
499 | ||
500 | return 1; | |
501 | } | |
502 | ||
503 | void __init cec_init(void) | |
504 | { | |
505 | if (ce_arr.disabled) | |
506 | return; | |
507 | ||
508 | ce_arr.array = (void *)get_zeroed_page(GFP_KERNEL); | |
509 | if (!ce_arr.array) { | |
510 | pr_err("Error allocating CE array page!\n"); | |
511 | return; | |
512 | } | |
513 | ||
514 | if (create_debugfs_nodes()) | |
515 | return; | |
516 | ||
2328874d CW |
517 | INIT_DELAYED_WORK(&cec_work, cec_work_fn); |
518 | schedule_delayed_work(&cec_work, CEC_DECAY_DEFAULT_INTERVAL); | |
011d8261 BP |
519 | |
520 | pr_info("Correctable Errors collector initialized.\n"); | |
521 | } | |
522 | ||
523 | int __init parse_cec_param(char *str) | |
524 | { | |
525 | if (!str) | |
526 | return 0; | |
527 | ||
528 | if (*str == '=') | |
529 | str++; | |
530 | ||
69a33000 | 531 | if (!strcmp(str, "cec_disable")) |
011d8261 BP |
532 | ce_arr.disabled = 1; |
533 | else | |
534 | return 0; | |
535 | ||
536 | return 1; | |
537 | } |