Borislav Petkov | 011d826 | 2017-03-27 11:33:02 +0200 | [diff] [blame] | 1 | #include <linux/mm.h> |
| 2 | #include <linux/gfp.h> |
| 3 | #include <linux/kernel.h> |
| 4 | |
| 5 | #include <asm/mce.h> |
| 6 | |
| 7 | #include "debugfs.h" |
| 8 | |
| 9 | /* |
| 10 | * RAS Correctable Errors Collector |
| 11 | * |
| 12 | * This is a simple gadget which collects correctable errors and counts their |
| 13 | * occurrence per physical page address. |
| 14 | * |
| 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 |
| 17 | * structure: |
| 18 | * |
| 19 | * [63 ... PFN ... 12 | 11 ... generation ... 10 | 9 ... count ... 0] |
| 20 | * |
| 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 |
| 24 | * then 00b. |
| 25 | * |
| 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. |
| 30 | * |
| 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 |
| 36 | * in the page. |
| 37 | * |
| 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. |
| 42 | * |
| 43 | * That error event entry causes cec_add_elem() to return !0 value and thus |
| 44 | * signal to its callers to log the error. |
| 45 | * |
| 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. |
| 51 | * |
| 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 |
| 54 | * patterns. |
| 55 | */ |
| 56 | |
| 57 | #undef pr_fmt |
| 58 | #define pr_fmt(fmt) "RAS: " fmt |
| 59 | |
| 60 | /* |
| 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. |
| 63 | */ |
| 64 | #define DECAY_BITS 2 |
| 65 | #define DECAY_MASK ((1ULL << DECAY_BITS) - 1) |
| 66 | #define MAX_ELEMS (PAGE_SIZE / sizeof(u64)) |
| 67 | |
| 68 | /* |
| 69 | * Threshold amount of inserted elements after which we start spring |
| 70 | * cleaning. |
| 71 | */ |
| 72 | #define CLEAN_ELEMS (MAX_ELEMS >> DECAY_BITS) |
| 73 | |
| 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) |
| 78 | |
| 79 | /* |
| 80 | * u64: [ 63 ... 12 | DECAY_BITS | COUNT_BITS ] |
| 81 | */ |
| 82 | |
| 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)) |
| 87 | |
| 88 | static struct ce_array { |
| 89 | u64 *array; /* container page */ |
| 90 | unsigned int n; /* number of elements in the array */ |
| 91 | |
| 92 | unsigned int decay_count; /* |
| 93 | * number of element insertions/increments |
| 94 | * since the last spring cleaning. |
| 95 | */ |
| 96 | |
| 97 | u64 pfns_poisoned; /* |
| 98 | * number of PFNs which got poisoned. |
| 99 | */ |
| 100 | |
| 101 | u64 ces_entered; /* |
| 102 | * The number of correctable errors |
| 103 | * entered into the collector. |
| 104 | */ |
| 105 | |
| 106 | u64 decays_done; /* |
| 107 | * Times we did spring cleaning. |
| 108 | */ |
| 109 | |
| 110 | union { |
| 111 | struct { |
| 112 | __u32 disabled : 1, /* cmdline disabled */ |
| 113 | __resv : 31; |
| 114 | }; |
| 115 | __u32 flags; |
| 116 | }; |
| 117 | } ce_arr; |
| 118 | |
| 119 | static DEFINE_MUTEX(ce_mutex); |
| 120 | static u64 dfs_pfn; |
| 121 | |
| 122 | /* Amount of errors after which we offline */ |
| 123 | static unsigned int count_threshold = COUNT_MASK; |
| 124 | |
| 125 | /* |
| 126 | * The timer "decays" element count each timer_interval which is 24hrs by |
| 127 | * default. |
| 128 | */ |
| 129 | |
| 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; |
| 135 | |
| 136 | /* |
| 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. |
| 139 | */ |
| 140 | static void do_spring_cleaning(struct ce_array *ca) |
| 141 | { |
| 142 | int i; |
| 143 | |
| 144 | for (i = 0; i < ca->n; i++) { |
| 145 | u8 decay = DECAY(ca->array[i]); |
| 146 | |
| 147 | if (!decay) |
| 148 | continue; |
| 149 | |
| 150 | decay--; |
| 151 | |
| 152 | ca->array[i] &= ~(DECAY_MASK << COUNT_BITS); |
| 153 | ca->array[i] |= (decay << COUNT_BITS); |
| 154 | } |
| 155 | ca->decay_count = 0; |
| 156 | ca->decays_done++; |
| 157 | } |
| 158 | |
| 159 | /* |
| 160 | * @interval in seconds |
| 161 | */ |
| 162 | static void cec_mod_timer(struct timer_list *t, unsigned long interval) |
| 163 | { |
| 164 | unsigned long iv; |
| 165 | |
| 166 | iv = interval * HZ + jiffies; |
| 167 | |
| 168 | mod_timer(t, round_jiffies(iv)); |
| 169 | } |
| 170 | |
| 171 | static void cec_timer_fn(unsigned long data) |
| 172 | { |
| 173 | struct ce_array *ca = (struct ce_array *)data; |
| 174 | |
| 175 | do_spring_cleaning(ca); |
| 176 | |
| 177 | cec_mod_timer(&cec_timer, timer_interval); |
| 178 | } |
| 179 | |
| 180 | /* |
| 181 | * @to: index of the smallest element which is >= then @pfn. |
| 182 | * |
| 183 | * Return the index of the pfn if found, otherwise negative value. |
| 184 | */ |
| 185 | static int __find_elem(struct ce_array *ca, u64 pfn, unsigned int *to) |
| 186 | { |
| 187 | u64 this_pfn; |
| 188 | int min = 0, max = ca->n; |
| 189 | |
| 190 | while (min < max) { |
| 191 | int tmp = (max + min) >> 1; |
| 192 | |
| 193 | this_pfn = PFN(ca->array[tmp]); |
| 194 | |
| 195 | if (this_pfn < pfn) |
| 196 | min = tmp + 1; |
| 197 | else if (this_pfn > pfn) |
| 198 | max = tmp; |
| 199 | else { |
| 200 | min = tmp; |
| 201 | break; |
| 202 | } |
| 203 | } |
| 204 | |
| 205 | if (to) |
| 206 | *to = min; |
| 207 | |
| 208 | this_pfn = PFN(ca->array[min]); |
| 209 | |
| 210 | if (this_pfn == pfn) |
| 211 | return min; |
| 212 | |
| 213 | return -ENOKEY; |
| 214 | } |
| 215 | |
| 216 | static int find_elem(struct ce_array *ca, u64 pfn, unsigned int *to) |
| 217 | { |
| 218 | WARN_ON(!to); |
| 219 | |
| 220 | if (!ca->n) { |
| 221 | *to = 0; |
| 222 | return -ENOKEY; |
| 223 | } |
| 224 | return __find_elem(ca, pfn, to); |
| 225 | } |
| 226 | |
| 227 | static void del_elem(struct ce_array *ca, int idx) |
| 228 | { |
| 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)); |
| 234 | |
| 235 | ca->n--; |
| 236 | } |
| 237 | |
| 238 | static u64 del_lru_elem_unlocked(struct ce_array *ca) |
| 239 | { |
| 240 | unsigned int min = FULL_COUNT_MASK; |
| 241 | int i, min_idx = 0; |
| 242 | |
| 243 | for (i = 0; i < ca->n; i++) { |
| 244 | unsigned int this = FULL_COUNT(ca->array[i]); |
| 245 | |
| 246 | if (min > this) { |
| 247 | min = this; |
| 248 | min_idx = i; |
| 249 | } |
| 250 | } |
| 251 | |
| 252 | del_elem(ca, min_idx); |
| 253 | |
| 254 | return PFN(ca->array[min_idx]); |
| 255 | } |
| 256 | |
| 257 | /* |
| 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. |
| 260 | */ |
| 261 | static u64 __maybe_unused del_lru_elem(void) |
| 262 | { |
| 263 | struct ce_array *ca = &ce_arr; |
| 264 | u64 pfn; |
| 265 | |
| 266 | if (!ca->n) |
| 267 | return 0; |
| 268 | |
| 269 | mutex_lock(&ce_mutex); |
| 270 | pfn = del_lru_elem_unlocked(ca); |
| 271 | mutex_unlock(&ce_mutex); |
| 272 | |
| 273 | return pfn; |
| 274 | } |
| 275 | |
| 276 | |
| 277 | int cec_add_elem(u64 pfn) |
| 278 | { |
| 279 | struct ce_array *ca = &ce_arr; |
| 280 | unsigned int to; |
| 281 | int count, ret = 0; |
| 282 | |
| 283 | /* |
| 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. |
| 286 | */ |
| 287 | if (!ce_arr.array || ce_arr.disabled) |
| 288 | return -ENODEV; |
| 289 | |
| 290 | ca->ces_entered++; |
| 291 | |
| 292 | mutex_lock(&ce_mutex); |
| 293 | |
| 294 | if (ca->n == MAX_ELEMS) |
| 295 | WARN_ON(!del_lru_elem_unlocked(ca)); |
| 296 | |
| 297 | ret = find_elem(ca, pfn, &to); |
| 298 | if (ret < 0) { |
| 299 | /* |
| 300 | * Shift range [to-end] to make room for one more element. |
| 301 | */ |
| 302 | memmove((void *)&ca->array[to + 1], |
| 303 | (void *)&ca->array[to], |
| 304 | (ca->n - to) * sizeof(u64)); |
| 305 | |
| 306 | ca->array[to] = (pfn << PAGE_SHIFT) | |
| 307 | (DECAY_MASK << COUNT_BITS) | 1; |
| 308 | |
| 309 | ca->n++; |
| 310 | |
| 311 | ret = 0; |
| 312 | |
| 313 | goto decay; |
| 314 | } |
| 315 | |
| 316 | count = COUNT(ca->array[to]); |
| 317 | |
| 318 | if (count < count_threshold) { |
| 319 | ca->array[to] |= (DECAY_MASK << COUNT_BITS); |
| 320 | ca->array[to]++; |
| 321 | |
| 322 | ret = 0; |
| 323 | } else { |
| 324 | u64 pfn = ca->array[to] >> PAGE_SHIFT; |
| 325 | |
| 326 | if (!pfn_valid(pfn)) { |
| 327 | pr_warn("CEC: Invalid pfn: 0x%llx\n", pfn); |
| 328 | } else { |
| 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); |
| 332 | ca->pfns_poisoned++; |
| 333 | } |
| 334 | |
| 335 | del_elem(ca, to); |
| 336 | |
| 337 | /* |
| 338 | * Return a >0 value to denote that we've reached the offlining |
| 339 | * threshold. |
| 340 | */ |
| 341 | ret = 1; |
| 342 | |
| 343 | goto unlock; |
| 344 | } |
| 345 | |
| 346 | decay: |
| 347 | ca->decay_count++; |
| 348 | |
| 349 | if (ca->decay_count >= CLEAN_ELEMS) |
| 350 | do_spring_cleaning(ca); |
| 351 | |
| 352 | unlock: |
| 353 | mutex_unlock(&ce_mutex); |
| 354 | |
| 355 | return ret; |
| 356 | } |
| 357 | |
| 358 | static int u64_get(void *data, u64 *val) |
| 359 | { |
| 360 | *val = *(u64 *)data; |
| 361 | |
| 362 | return 0; |
| 363 | } |
| 364 | |
| 365 | static int pfn_set(void *data, u64 val) |
| 366 | { |
| 367 | *(u64 *)data = val; |
| 368 | |
| 369 | return cec_add_elem(val); |
| 370 | } |
| 371 | |
| 372 | DEFINE_DEBUGFS_ATTRIBUTE(pfn_ops, u64_get, pfn_set, "0x%llx\n"); |
| 373 | |
| 374 | static int decay_interval_set(void *data, u64 val) |
| 375 | { |
| 376 | *(u64 *)data = val; |
| 377 | |
| 378 | if (val < CEC_TIMER_MIN_INTERVAL) |
| 379 | return -EINVAL; |
| 380 | |
| 381 | if (val > CEC_TIMER_MAX_INTERVAL) |
| 382 | return -EINVAL; |
| 383 | |
| 384 | timer_interval = val; |
| 385 | |
| 386 | cec_mod_timer(&cec_timer, timer_interval); |
| 387 | return 0; |
| 388 | } |
| 389 | DEFINE_DEBUGFS_ATTRIBUTE(decay_interval_ops, u64_get, decay_interval_set, "%lld\n"); |
| 390 | |
| 391 | static int count_threshold_set(void *data, u64 val) |
| 392 | { |
| 393 | *(u64 *)data = val; |
| 394 | |
| 395 | if (val > COUNT_MASK) |
| 396 | val = COUNT_MASK; |
| 397 | |
| 398 | count_threshold = val; |
| 399 | |
| 400 | return 0; |
| 401 | } |
| 402 | DEFINE_DEBUGFS_ATTRIBUTE(count_threshold_ops, u64_get, count_threshold_set, "%lld\n"); |
| 403 | |
| 404 | static int array_dump(struct seq_file *m, void *v) |
| 405 | { |
| 406 | struct ce_array *ca = &ce_arr; |
| 407 | u64 prev = 0; |
| 408 | int i; |
| 409 | |
| 410 | mutex_lock(&ce_mutex); |
| 411 | |
| 412 | seq_printf(m, "{ n: %d\n", ca->n); |
| 413 | for (i = 0; i < ca->n; i++) { |
| 414 | u64 this = PFN(ca->array[i]); |
| 415 | |
| 416 | seq_printf(m, " %03d: [%016llx|%03llx]\n", i, this, FULL_COUNT(ca->array[i])); |
| 417 | |
| 418 | WARN_ON(prev > this); |
| 419 | |
| 420 | prev = this; |
| 421 | } |
| 422 | |
| 423 | seq_printf(m, "}\n"); |
| 424 | |
| 425 | seq_printf(m, "Stats:\nCEs: %llu\nofflined pages: %llu\n", |
| 426 | ca->ces_entered, ca->pfns_poisoned); |
| 427 | |
| 428 | seq_printf(m, "Flags: 0x%x\n", ca->flags); |
| 429 | |
| 430 | seq_printf(m, "Timer interval: %lld seconds\n", timer_interval); |
| 431 | seq_printf(m, "Decays: %lld\n", ca->decays_done); |
| 432 | |
| 433 | seq_printf(m, "Action threshold: %d\n", count_threshold); |
| 434 | |
| 435 | mutex_unlock(&ce_mutex); |
| 436 | |
| 437 | return 0; |
| 438 | } |
| 439 | |
| 440 | static int array_open(struct inode *inode, struct file *filp) |
| 441 | { |
| 442 | return single_open(filp, array_dump, NULL); |
| 443 | } |
| 444 | |
| 445 | static const struct file_operations array_ops = { |
| 446 | .owner = THIS_MODULE, |
| 447 | .open = array_open, |
| 448 | .read = seq_read, |
| 449 | .llseek = seq_lseek, |
| 450 | .release = single_release, |
| 451 | }; |
| 452 | |
| 453 | static int __init create_debugfs_nodes(void) |
| 454 | { |
| 455 | struct dentry *d, *pfn, *decay, *count, *array; |
| 456 | |
| 457 | d = debugfs_create_dir("cec", ras_debugfs_dir); |
| 458 | if (!d) { |
| 459 | pr_warn("Error creating cec debugfs node!\n"); |
| 460 | return -1; |
| 461 | } |
| 462 | |
| 463 | pfn = debugfs_create_file("pfn", S_IRUSR | S_IWUSR, d, &dfs_pfn, &pfn_ops); |
| 464 | if (!pfn) { |
| 465 | pr_warn("Error creating pfn debugfs node!\n"); |
| 466 | goto err; |
| 467 | } |
| 468 | |
| 469 | array = debugfs_create_file("array", S_IRUSR, d, NULL, &array_ops); |
| 470 | if (!array) { |
| 471 | pr_warn("Error creating array debugfs node!\n"); |
| 472 | goto err; |
| 473 | } |
| 474 | |
| 475 | decay = debugfs_create_file("decay_interval", S_IRUSR | S_IWUSR, d, |
| 476 | &timer_interval, &decay_interval_ops); |
| 477 | if (!decay) { |
| 478 | pr_warn("Error creating decay_interval debugfs node!\n"); |
| 479 | goto err; |
| 480 | } |
| 481 | |
| 482 | count = debugfs_create_file("count_threshold", S_IRUSR | S_IWUSR, d, |
| 483 | &count_threshold, &count_threshold_ops); |
| 484 | if (!decay) { |
| 485 | pr_warn("Error creating count_threshold debugfs node!\n"); |
| 486 | goto err; |
| 487 | } |
| 488 | |
| 489 | |
| 490 | return 0; |
| 491 | |
| 492 | err: |
| 493 | debugfs_remove_recursive(d); |
| 494 | |
| 495 | return 1; |
| 496 | } |
| 497 | |
| 498 | void __init cec_init(void) |
| 499 | { |
| 500 | if (ce_arr.disabled) |
| 501 | return; |
| 502 | |
| 503 | ce_arr.array = (void *)get_zeroed_page(GFP_KERNEL); |
| 504 | if (!ce_arr.array) { |
| 505 | pr_err("Error allocating CE array page!\n"); |
| 506 | return; |
| 507 | } |
| 508 | |
| 509 | if (create_debugfs_nodes()) |
| 510 | return; |
| 511 | |
| 512 | setup_timer(&cec_timer, cec_timer_fn, (unsigned long)&ce_arr); |
| 513 | cec_mod_timer(&cec_timer, CEC_TIMER_DEFAULT_INTERVAL); |
| 514 | |
| 515 | pr_info("Correctable Errors collector initialized.\n"); |
| 516 | } |
| 517 | |
| 518 | int __init parse_cec_param(char *str) |
| 519 | { |
| 520 | if (!str) |
| 521 | return 0; |
| 522 | |
| 523 | if (*str == '=') |
| 524 | str++; |
| 525 | |
| 526 | if (!strncmp(str, "cec_disable", 7)) |
| 527 | ce_arr.disabled = 1; |
| 528 | else |
| 529 | return 0; |
| 530 | |
| 531 | return 1; |
| 532 | } |