| 1 |
|
| 2 | |
| 3 | * edac_device.c
|
| 4 | * (C) 2007 www.douglaskthompson.com
|
| 5 | *
|
| 6 | * This file may be distributed under the terms of the
|
| 7 | * GNU General Public License.
|
| 8 | *
|
| 9 | * Written by Doug Thompson <[email protected]>
|
| 10 | *
|
| 11 | * edac_device API implementation
|
| 12 | * 19 Jan 2007
|
| 13 |
|
| 14 |
|
| 15 | #include <asm/page.h>
|
| 16 | #include <linux/uaccess.h>
|
| 17 | #include <linux/ctype.h>
|
| 18 | #include <linux/highmem.h>
|
| 19 | #include <linux/init.h>
|
| 20 | #include <linux/jiffies.h>
|
| 21 | #include <linux/module.h>
|
| 22 | #include <linux/slab.h>
|
| 23 | #include <linux/smp.h>
|
| 24 | #include <linux/spinlock.h>
|
| 25 | #include <linux/sysctl.h>
|
| 26 | #include <linux/timer.h>
|
| 27 |
|
| 28 | #include "edac_device.h"
|
| 29 | #include "edac_module.h"
|
| 30 |
|
| 31 | |
| 32 | * is protected by the 'device_ctls_mutex' lock
|
| 33 |
|
| 34 | static DEFINE_MUTEX(device_ctls_mutex);
|
| 35 | static LIST_HEAD(edac_device_list);
|
| 36 |
|
| 37 |
|
| 38 | #define DEFAULT_POLL_INTERVAL 1000
|
| 39 |
|
| 40 | #ifdef CONFIG_EDAC_DEBUG
|
| 41 | static void edac_device_dump_device(struct edac_device_ctl_info *edac_dev)
|
| 42 | {
|
| 43 | edac_dbg(3, "\tedac_dev = %p dev_idx=%d\n",
|
| 44 | edac_dev, edac_dev->dev_idx);
|
| 45 | edac_dbg(4, "\tedac_dev->edac_check = %p\n", edac_dev->edac_check);
|
| 46 | edac_dbg(3, "\tdev = %p\n", edac_dev->dev);
|
| 47 | edac_dbg(3, "\tmod_name:ctl_name = %s:%s\n",
|
| 48 | edac_dev->mod_name, edac_dev->ctl_name);
|
| 49 | edac_dbg(3, "\tpvt_info = %p\n\n", edac_dev->pvt_info);
|
| 50 | }
|
| 51 | #endif
|
| 52 |
|
| 53 | |
| 54 | * @off_val: zero, 1, or other based offset
|
| 55 |
|
| 56 | struct edac_device_ctl_info *
|
| 57 | edac_device_alloc_ctl_info(unsigned pvt_sz, char *dev_name, unsigned nr_instances,
|
| 58 | char *blk_name, unsigned nr_blocks, unsigned off_val,
|
| 59 | int device_index)
|
| 60 | {
|
| 61 | struct edac_device_block *dev_blk, *blk_p, *blk;
|
| 62 | struct edac_device_instance *dev_inst, *inst;
|
| 63 | struct edac_device_ctl_info *dev_ctl;
|
| 64 | unsigned instance, block;
|
| 65 | void *pvt;
|
| 66 | int err;
|
| 67 |
|
| 68 | edac_dbg(4, "instances=%d blocks=%d\n", nr_instances, nr_blocks);
|
| 69 |
|
| 70 | dev_ctl = kzalloc_obj(struct edac_device_ctl_info);
|
| 71 | if (!dev_ctl)
|
| 72 | return NULL;
|
| 73 |
|
| 74 | dev_inst = kzalloc_objs(struct edac_device_instance, nr_instances);
|
| 75 | if (!dev_inst)
|
| 76 | goto free;
|
| 77 |
|
| 78 | dev_ctl->instances = dev_inst;
|
| 79 |
|
| 80 | dev_blk = kzalloc_objs(struct edac_device_block,
|
| 81 | nr_instances * nr_blocks);
|
| 82 | if (!dev_blk)
|
| 83 | goto free;
|
| 84 |
|
| 85 | dev_ctl->blocks = dev_blk;
|
| 86 |
|
| 87 | if (pvt_sz) {
|
| 88 | pvt = kzalloc(pvt_sz, GFP_KERNEL);
|
| 89 | if (!pvt)
|
| 90 | goto free;
|
| 91 |
|
| 92 | dev_ctl->pvt_info = pvt;
|
| 93 | }
|
| 94 |
|
| 95 | dev_ctl->dev_idx = device_index;
|
| 96 | dev_ctl->nr_instances = nr_instances;
|
| 97 |
|
| 98 |
|
| 99 | dev_ctl->log_ce = 1;
|
| 100 | dev_ctl->log_ue = 1;
|
| 101 |
|
| 102 |
|
| 103 | snprintf(dev_ctl->name, sizeof(dev_ctl->name),"%s", dev_name);
|
| 104 |
|
| 105 |
|
| 106 | for (instance = 0; instance < nr_instances; instance++) {
|
| 107 | inst = &dev_inst[instance];
|
| 108 | inst->ctl = dev_ctl;
|
| 109 | inst->nr_blocks = nr_blocks;
|
| 110 | blk_p = &dev_blk[instance * nr_blocks];
|
| 111 | inst->blocks = blk_p;
|
| 112 |
|
| 113 |
|
| 114 | snprintf(inst->name, sizeof(inst->name), "%s%u", dev_name, instance);
|
| 115 |
|
| 116 |
|
| 117 | for (block = 0; block < nr_blocks; block++) {
|
| 118 | blk = &blk_p[block];
|
| 119 | blk->instance = inst;
|
| 120 | snprintf(blk->name, sizeof(blk->name),
|
| 121 | "%s%d", blk_name, block + off_val);
|
| 122 |
|
| 123 | edac_dbg(4, "instance=%d inst_p=%p block=#%d block_p=%p name='%s'\n",
|
| 124 | instance, inst, block, blk, blk->name);
|
| 125 | }
|
| 126 | }
|
| 127 |
|
| 128 |
|
| 129 | dev_ctl->op_state = OP_ALLOC;
|
| 130 |
|
| 131 | |
| 132 | * Initialize the 'root' kobj for the edac_device controller
|
| 133 |
|
| 134 | err = edac_device_register_sysfs_main_kobj(dev_ctl);
|
| 135 | if (err)
|
| 136 | goto free;
|
| 137 |
|
| 138 | |
| 139 | * 'free' the object, then the function:
|
| 140 | * edac_device_unregister_sysfs_main_kobj() must be called
|
| 141 | * which will perform kobj unregistration and the actual free
|
| 142 | * will occur during the kobject callback operation
|
| 143 |
|
| 144 |
|
| 145 | return dev_ctl;
|
| 146 |
|
| 147 | free:
|
| 148 | __edac_device_free_ctl_info(dev_ctl);
|
| 149 |
|
| 150 | return NULL;
|
| 151 | }
|
| 152 | EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
|
| 153 |
|
| 154 | void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)
|
| 155 | {
|
| 156 | edac_device_unregister_sysfs_main_kobj(ctl_info);
|
| 157 | }
|
| 158 | EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
|
| 159 |
|
| 160 | |
| 161 | * find_edac_device_by_dev
|
| 162 | * scans the edac_device list for a specific 'struct device *'
|
| 163 | *
|
| 164 | * lock to be held prior to call: device_ctls_mutex
|
| 165 | *
|
| 166 | * Return:
|
| 167 | * pointer to control structure managing 'dev'
|
| 168 | * NULL if not found on list
|
| 169 |
|
| 170 | static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
|
| 171 | {
|
| 172 | struct edac_device_ctl_info *edac_dev;
|
| 173 | struct list_head *item;
|
| 174 |
|
| 175 | edac_dbg(0, "\n");
|
| 176 |
|
| 177 | list_for_each(item, &edac_device_list) {
|
| 178 | edac_dev = list_entry(item, struct edac_device_ctl_info, link);
|
| 179 |
|
| 180 | if (edac_dev->dev == dev)
|
| 181 | return edac_dev;
|
| 182 | }
|
| 183 |
|
| 184 | return NULL;
|
| 185 | }
|
| 186 |
|
| 187 | |
| 188 | * add_edac_dev_to_global_list
|
| 189 | * Before calling this function, caller must
|
| 190 | * assign a unique value to edac_dev->dev_idx.
|
| 191 | *
|
| 192 | * lock to be held prior to call: device_ctls_mutex
|
| 193 | *
|
| 194 | * Return:
|
| 195 | * 0 on success
|
| 196 | * 1 on failure.
|
| 197 |
|
| 198 | static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
|
| 199 | {
|
| 200 | struct list_head *item, *insert_before;
|
| 201 | struct edac_device_ctl_info *rover;
|
| 202 |
|
| 203 | insert_before = &edac_device_list;
|
| 204 |
|
| 205 |
|
| 206 | rover = find_edac_device_by_dev(edac_dev->dev);
|
| 207 | if (unlikely(rover != NULL))
|
| 208 | goto fail0;
|
| 209 |
|
| 210 |
|
| 211 | list_for_each(item, &edac_device_list) {
|
| 212 | rover = list_entry(item, struct edac_device_ctl_info, link);
|
| 213 |
|
| 214 | if (rover->dev_idx >= edac_dev->dev_idx) {
|
| 215 | if (unlikely(rover->dev_idx == edac_dev->dev_idx))
|
| 216 | goto fail1;
|
| 217 |
|
| 218 | insert_before = item;
|
| 219 | break;
|
| 220 | }
|
| 221 | }
|
| 222 |
|
| 223 | list_add_tail_rcu(&edac_dev->link, insert_before);
|
| 224 | return 0;
|
| 225 |
|
| 226 | fail0:
|
| 227 | edac_printk(KERN_WARNING, EDAC_MC,
|
| 228 | "%s (%s) %s %s already assigned %d\n",
|
| 229 | dev_name(rover->dev), edac_dev_name(rover),
|
| 230 | rover->mod_name, rover->ctl_name, rover->dev_idx);
|
| 231 | return 1;
|
| 232 |
|
| 233 | fail1:
|
| 234 | edac_printk(KERN_WARNING, EDAC_MC,
|
| 235 | "bug in low-level driver: attempt to assign\n"
|
| 236 | " duplicate dev_idx %d in %s()\n", rover->dev_idx,
|
| 237 | __func__);
|
| 238 | return 1;
|
| 239 | }
|
| 240 |
|
| 241 | |
| 242 | * del_edac_device_from_global_list
|
| 243 |
|
| 244 | static void del_edac_device_from_global_list(struct edac_device_ctl_info
|
| 245 | *edac_device)
|
| 246 | {
|
| 247 | list_del_rcu(&edac_device->link);
|
| 248 |
|
| 249 | |
| 250 | * NMI handlers may be traversing list
|
| 251 |
|
| 252 | synchronize_rcu();
|
| 253 | INIT_LIST_HEAD(&edac_device->link);
|
| 254 | }
|
| 255 |
|
| 256 | |
| 257 | * edac_device_workq_function
|
| 258 | * performs the operation scheduled by a workq request
|
| 259 | *
|
| 260 | * this workq is embedded within an edac_device_ctl_info
|
| 261 | * structure, that needs to be polled for possible error events.
|
| 262 | *
|
| 263 | * This operation is to acquire the list mutex lock
|
| 264 | * (thus preventing insertation or deletion)
|
| 265 | * and then call the device's poll function IFF this device is
|
| 266 | * running polled and there is a poll function defined.
|
| 267 |
|
| 268 | static void edac_device_workq_function(struct work_struct *work_req)
|
| 269 | {
|
| 270 | struct delayed_work *d_work = to_delayed_work(work_req);
|
| 271 | struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);
|
| 272 |
|
| 273 | mutex_lock(&device_ctls_mutex);
|
| 274 |
|
| 275 |
|
| 276 | if (edac_dev->op_state == OP_OFFLINE) {
|
| 277 | mutex_unlock(&device_ctls_mutex);
|
| 278 | return;
|
| 279 | }
|
| 280 |
|
| 281 |
|
| 282 | if ((edac_dev->op_state == OP_RUNNING_POLL) &&
|
| 283 | (edac_dev->edac_check != NULL)) {
|
| 284 | edac_dev->edac_check(edac_dev);
|
| 285 | }
|
| 286 |
|
| 287 | mutex_unlock(&device_ctls_mutex);
|
| 288 |
|
| 289 | |
| 290 | * if the number of msec is for 1 sec, then adjust to the next
|
| 291 | * whole one second to save timers firing all over the period
|
| 292 | * between integral seconds
|
| 293 |
|
| 294 | if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)
|
| 295 | edac_queue_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));
|
| 296 | else
|
| 297 | edac_queue_work(&edac_dev->work, edac_dev->delay);
|
| 298 | }
|
| 299 |
|
| 300 | |
| 301 | * edac_device_workq_setup
|
| 302 | * initialize a workq item for this edac_device instance
|
| 303 | * passing in the new delay period in msec
|
| 304 |
|
| 305 | static void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
|
| 306 | unsigned msec)
|
| 307 | {
|
| 308 | edac_dbg(0, "\n");
|
| 309 |
|
| 310 | |
| 311 | * to used in the time period calculation
|
| 312 | * then calc the number of jiffies that represents
|
| 313 |
|
| 314 | edac_dev->poll_msec = msec;
|
| 315 | edac_dev->delay = msecs_to_jiffies(msec);
|
| 316 |
|
| 317 | INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
|
| 318 |
|
| 319 | |
| 320 | * fire ON the 1 second time event. This helps reduce all sorts of
|
| 321 | * timers firing on sub-second basis, while they are happy
|
| 322 | * to fire together on the 1 second exactly
|
| 323 |
|
| 324 | if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)
|
| 325 | edac_queue_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));
|
| 326 | else
|
| 327 | edac_queue_work(&edac_dev->work, edac_dev->delay);
|
| 328 | }
|
| 329 |
|
| 330 | |
| 331 | * edac_device_workq_teardown
|
| 332 | * stop the workq processing on this edac_dev
|
| 333 |
|
| 334 | static void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)
|
| 335 | {
|
| 336 | if (!edac_dev->edac_check)
|
| 337 | return;
|
| 338 |
|
| 339 | edac_dev->op_state = OP_OFFLINE;
|
| 340 |
|
| 341 | edac_stop_work(&edac_dev->work);
|
| 342 | }
|
| 343 |
|
| 344 | |
| 345 | * Stop any outstanding workq queued up at this time because sleep time will
|
| 346 | * be reset. Then restart the workq on the new delay.
|
| 347 |
|
| 348 | void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev, unsigned int msec)
|
| 349 | {
|
| 350 | edac_dev->poll_msec = msec;
|
| 351 | edac_dev->delay = msecs_to_jiffies(msec);
|
| 352 |
|
| 353 |
|
| 354 | if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)
|
| 355 | edac_mod_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));
|
| 356 | else
|
| 357 | edac_mod_work(&edac_dev->work, edac_dev->delay);
|
| 358 | }
|
| 359 |
|
| 360 | int edac_device_alloc_index(void)
|
| 361 | {
|
| 362 | static atomic_t device_indexes = ATOMIC_INIT(0);
|
| 363 |
|
| 364 | return atomic_inc_return(&device_indexes) - 1;
|
| 365 | }
|
| 366 | EXPORT_SYMBOL_GPL(edac_device_alloc_index);
|
| 367 |
|
| 368 | int edac_device_add_device(struct edac_device_ctl_info *edac_dev)
|
| 369 | {
|
| 370 | edac_dbg(0, "\n");
|
| 371 |
|
| 372 | #ifdef CONFIG_EDAC_DEBUG
|
| 373 | if (edac_debug_level >= 3)
|
| 374 | edac_device_dump_device(edac_dev);
|
| 375 | #endif
|
| 376 | mutex_lock(&device_ctls_mutex);
|
| 377 |
|
| 378 | if (add_edac_dev_to_global_list(edac_dev))
|
| 379 | goto fail0;
|
| 380 |
|
| 381 |
|
| 382 | edac_dev->start_time = jiffies;
|
| 383 |
|
| 384 |
|
| 385 | if (edac_device_create_sysfs(edac_dev)) {
|
| 386 | edac_device_printk(edac_dev, KERN_WARNING,
|
| 387 | "failed to create sysfs device\n");
|
| 388 | goto fail1;
|
| 389 | }
|
| 390 |
|
| 391 |
|
| 392 | if (edac_dev->edac_check != NULL) {
|
| 393 |
|
| 394 | edac_dev->op_state = OP_RUNNING_POLL;
|
| 395 |
|
| 396 | edac_device_workq_setup(edac_dev, edac_dev->poll_msec ?: DEFAULT_POLL_INTERVAL);
|
| 397 | } else {
|
| 398 | edac_dev->op_state = OP_RUNNING_INTERRUPT;
|
| 399 | }
|
| 400 |
|
| 401 |
|
| 402 | edac_device_printk(edac_dev, KERN_INFO,
|
| 403 | "Giving out device to module %s controller %s: DEV %s (%s)\n",
|
| 404 | edac_dev->mod_name, edac_dev->ctl_name, edac_dev->dev_name,
|
| 405 | edac_op_state_to_string(edac_dev->op_state));
|
| 406 |
|
| 407 | mutex_unlock(&device_ctls_mutex);
|
| 408 | return 0;
|
| 409 |
|
| 410 | fail1:
|
| 411 |
|
| 412 | del_edac_device_from_global_list(edac_dev);
|
| 413 |
|
| 414 | fail0:
|
| 415 | mutex_unlock(&device_ctls_mutex);
|
| 416 | return 1;
|
| 417 | }
|
| 418 | EXPORT_SYMBOL_GPL(edac_device_add_device);
|
| 419 |
|
| 420 | struct edac_device_ctl_info *edac_device_del_device(struct device *dev)
|
| 421 | {
|
| 422 | struct edac_device_ctl_info *edac_dev;
|
| 423 |
|
| 424 | edac_dbg(0, "\n");
|
| 425 |
|
| 426 | mutex_lock(&device_ctls_mutex);
|
| 427 |
|
| 428 |
|
| 429 | edac_dev = find_edac_device_by_dev(dev);
|
| 430 | if (edac_dev == NULL) {
|
| 431 | mutex_unlock(&device_ctls_mutex);
|
| 432 | return NULL;
|
| 433 | }
|
| 434 |
|
| 435 |
|
| 436 | edac_dev->op_state = OP_OFFLINE;
|
| 437 |
|
| 438 |
|
| 439 | del_edac_device_from_global_list(edac_dev);
|
| 440 |
|
| 441 | mutex_unlock(&device_ctls_mutex);
|
| 442 |
|
| 443 |
|
| 444 | edac_device_workq_teardown(edac_dev);
|
| 445 |
|
| 446 |
|
| 447 | edac_device_remove_sysfs(edac_dev);
|
| 448 |
|
| 449 | edac_printk(KERN_INFO, EDAC_MC,
|
| 450 | "Removed device %d for %s %s: DEV %s\n",
|
| 451 | edac_dev->dev_idx,
|
| 452 | edac_dev->mod_name, edac_dev->ctl_name, edac_dev_name(edac_dev));
|
| 453 |
|
| 454 | return edac_dev;
|
| 455 | }
|
| 456 | EXPORT_SYMBOL_GPL(edac_device_del_device);
|
| 457 |
|
| 458 | static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
|
| 459 | {
|
| 460 | return edac_dev->log_ce;
|
| 461 | }
|
| 462 |
|
| 463 | static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
|
| 464 | {
|
| 465 | return edac_dev->log_ue;
|
| 466 | }
|
| 467 |
|
| 468 | static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
|
| 469 | *edac_dev)
|
| 470 | {
|
| 471 | return edac_dev->panic_on_ue;
|
| 472 | }
|
| 473 |
|
| 474 | void edac_device_handle_ce_count(struct edac_device_ctl_info *edac_dev,
|
| 475 | unsigned int count, int inst_nr, int block_nr,
|
| 476 | const char *msg)
|
| 477 | {
|
| 478 | struct edac_device_instance *instance;
|
| 479 | struct edac_device_block *block = NULL;
|
| 480 |
|
| 481 | if (!count)
|
| 482 | return;
|
| 483 |
|
| 484 | if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
|
| 485 | edac_device_printk(edac_dev, KERN_ERR,
|
| 486 | "INTERNAL ERROR: 'instance' out of range "
|
| 487 | "(%d >= %d)\n", inst_nr,
|
| 488 | edac_dev->nr_instances);
|
| 489 | return;
|
| 490 | }
|
| 491 |
|
| 492 | instance = edac_dev->instances + inst_nr;
|
| 493 |
|
| 494 | if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
|
| 495 | edac_device_printk(edac_dev, KERN_ERR,
|
| 496 | "INTERNAL ERROR: instance %d 'block' "
|
| 497 | "out of range (%d >= %d)\n",
|
| 498 | inst_nr, block_nr,
|
| 499 | instance->nr_blocks);
|
| 500 | return;
|
| 501 | }
|
| 502 |
|
| 503 | if (instance->nr_blocks > 0) {
|
| 504 | block = instance->blocks + block_nr;
|
| 505 | block->counters.ce_count += count;
|
| 506 | }
|
| 507 |
|
| 508 |
|
| 509 | instance->counters.ce_count += count;
|
| 510 | edac_dev->counters.ce_count += count;
|
| 511 |
|
| 512 | if (edac_device_get_log_ce(edac_dev))
|
| 513 | edac_device_printk(edac_dev, KERN_WARNING,
|
| 514 | "CE: %s instance: %s block: %s count: %d '%s'\n",
|
| 515 | edac_dev->ctl_name, instance->name,
|
| 516 | block ? block->name : "N/A", count, msg);
|
| 517 | }
|
| 518 | EXPORT_SYMBOL_GPL(edac_device_handle_ce_count);
|
| 519 |
|
| 520 | void edac_device_handle_ue_count(struct edac_device_ctl_info *edac_dev,
|
| 521 | unsigned int count, int inst_nr, int block_nr,
|
| 522 | const char *msg)
|
| 523 | {
|
| 524 | struct edac_device_instance *instance;
|
| 525 | struct edac_device_block *block = NULL;
|
| 526 |
|
| 527 | if (!count)
|
| 528 | return;
|
| 529 |
|
| 530 | if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
|
| 531 | edac_device_printk(edac_dev, KERN_ERR,
|
| 532 | "INTERNAL ERROR: 'instance' out of range "
|
| 533 | "(%d >= %d)\n", inst_nr,
|
| 534 | edac_dev->nr_instances);
|
| 535 | return;
|
| 536 | }
|
| 537 |
|
| 538 | instance = edac_dev->instances + inst_nr;
|
| 539 |
|
| 540 | if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
|
| 541 | edac_device_printk(edac_dev, KERN_ERR,
|
| 542 | "INTERNAL ERROR: instance %d 'block' "
|
| 543 | "out of range (%d >= %d)\n",
|
| 544 | inst_nr, block_nr,
|
| 545 | instance->nr_blocks);
|
| 546 | return;
|
| 547 | }
|
| 548 |
|
| 549 | if (instance->nr_blocks > 0) {
|
| 550 | block = instance->blocks + block_nr;
|
| 551 | block->counters.ue_count += count;
|
| 552 | }
|
| 553 |
|
| 554 |
|
| 555 | instance->counters.ue_count += count;
|
| 556 | edac_dev->counters.ue_count += count;
|
| 557 |
|
| 558 | if (edac_device_get_log_ue(edac_dev))
|
| 559 | edac_device_printk(edac_dev, KERN_EMERG,
|
| 560 | "UE: %s instance: %s block: %s count: %d '%s'\n",
|
| 561 | edac_dev->ctl_name, instance->name,
|
| 562 | block ? block->name : "N/A", count, msg);
|
| 563 |
|
| 564 | if (edac_device_get_panic_on_ue(edac_dev))
|
| 565 | panic("EDAC %s: UE instance: %s block %s count: %d '%s'\n",
|
| 566 | edac_dev->ctl_name, instance->name,
|
| 567 | block ? block->name : "N/A", count, msg);
|
| 568 | }
|
| 569 | EXPORT_SYMBOL_GPL(edac_device_handle_ue_count);
|
| 570 |
|
| 571 | static void edac_dev_release(struct device *dev)
|
| 572 | {
|
| 573 | struct edac_dev_feat_ctx *ctx = container_of(dev, struct edac_dev_feat_ctx, dev);
|
| 574 |
|
| 575 | kfree(ctx->mem_repair);
|
| 576 | kfree(ctx->scrub);
|
| 577 | kfree(ctx->dev.groups);
|
| 578 | kfree(ctx);
|
| 579 | }
|
| 580 |
|
| 581 | static const struct device_type edac_dev_type = {
|
| 582 | .name = "edac_dev",
|
| 583 | .release = edac_dev_release,
|
| 584 | };
|
| 585 |
|
| 586 | static void edac_dev_unreg(void *data)
|
| 587 | {
|
| 588 | device_unregister(data);
|
| 589 | }
|
| 590 |
|
| 591 | |
| 592 | * edac_dev_register - register device for RAS features with EDAC
|
| 593 | * @parent: parent device.
|
| 594 | * @name: name for the folder in the /sys/bus/edac/devices/,
|
| 595 | * which is derived from the parent device.
|
| 596 | * For e.g. /sys/bus/edac/devices/cxl_mem0/
|
| 597 | * @private: parent driver's data to store in the context if any.
|
| 598 | * @num_features: number of RAS features to register.
|
| 599 | * @ras_features: list of RAS features to register.
|
| 600 | *
|
| 601 | * Return:
|
| 602 | * * %0 - Success.
|
| 603 | * * %-EINVAL - Invalid parameters passed.
|
| 604 | * * %-ENOMEM - Dynamic memory allocation failed.
|
| 605 | *
|
| 606 |
|
| 607 | int edac_dev_register(struct device *parent, char *name,
|
| 608 | void *private, int num_features,
|
| 609 | const struct edac_dev_feature *ras_features)
|
| 610 | {
|
| 611 | const struct attribute_group **ras_attr_groups;
|
| 612 | struct edac_dev_data *dev_data;
|
| 613 | struct edac_dev_feat_ctx *ctx;
|
| 614 | int mem_repair_cnt = 0;
|
| 615 | int attr_gcnt = 0;
|
| 616 | int ret = -ENOMEM;
|
| 617 | int scrub_cnt = 0;
|
| 618 | int feat;
|
| 619 |
|
| 620 | if (!parent || !name || !num_features || !ras_features)
|
| 621 | return -EINVAL;
|
| 622 |
|
| 623 |
|
| 624 | for (feat = 0; feat < num_features; feat++) {
|
| 625 | switch (ras_features[feat].ft_type) {
|
| 626 | case RAS_FEAT_SCRUB:
|
| 627 | attr_gcnt++;
|
| 628 | scrub_cnt++;
|
| 629 | break;
|
| 630 | case RAS_FEAT_ECS:
|
| 631 | attr_gcnt += ras_features[feat].ecs_info.num_media_frus;
|
| 632 | break;
|
| 633 | case RAS_FEAT_MEM_REPAIR:
|
| 634 | attr_gcnt++;
|
| 635 | mem_repair_cnt++;
|
| 636 | break;
|
| 637 | default:
|
| 638 | return -EINVAL;
|
| 639 | }
|
| 640 | }
|
| 641 |
|
| 642 | ctx = kzalloc_obj(*ctx);
|
| 643 | if (!ctx)
|
| 644 | return -ENOMEM;
|
| 645 |
|
| 646 | ras_attr_groups = kzalloc_objs(*ras_attr_groups, attr_gcnt + 1);
|
| 647 | if (!ras_attr_groups)
|
| 648 | goto ctx_free;
|
| 649 |
|
| 650 | if (scrub_cnt) {
|
| 651 | ctx->scrub = kzalloc_objs(*ctx->scrub, scrub_cnt);
|
| 652 | if (!ctx->scrub)
|
| 653 | goto groups_free;
|
| 654 | }
|
| 655 |
|
| 656 | if (mem_repair_cnt) {
|
| 657 | ctx->mem_repair = kzalloc_objs(*ctx->mem_repair, mem_repair_cnt);
|
| 658 | if (!ctx->mem_repair)
|
| 659 | goto data_mem_free;
|
| 660 | }
|
| 661 |
|
| 662 | attr_gcnt = 0;
|
| 663 | scrub_cnt = 0;
|
| 664 | mem_repair_cnt = 0;
|
| 665 | for (feat = 0; feat < num_features; feat++, ras_features++) {
|
| 666 | switch (ras_features->ft_type) {
|
| 667 | case RAS_FEAT_SCRUB:
|
| 668 | if (!ras_features->scrub_ops || scrub_cnt != ras_features->instance) {
|
| 669 | ret = -EINVAL;
|
| 670 | goto data_mem_free;
|
| 671 | }
|
| 672 |
|
| 673 | dev_data = &ctx->scrub[scrub_cnt];
|
| 674 | dev_data->instance = scrub_cnt;
|
| 675 | dev_data->scrub_ops = ras_features->scrub_ops;
|
| 676 | dev_data->private = ras_features->ctx;
|
| 677 | ret = edac_scrub_get_desc(parent, &ras_attr_groups[attr_gcnt],
|
| 678 | ras_features->instance);
|
| 679 | if (ret)
|
| 680 | goto data_mem_free;
|
| 681 |
|
| 682 | scrub_cnt++;
|
| 683 | attr_gcnt++;
|
| 684 | break;
|
| 685 | case RAS_FEAT_ECS:
|
| 686 | if (!ras_features->ecs_ops) {
|
| 687 | ret = -EINVAL;
|
| 688 | goto data_mem_free;
|
| 689 | }
|
| 690 |
|
| 691 | dev_data = &ctx->ecs;
|
| 692 | dev_data->ecs_ops = ras_features->ecs_ops;
|
| 693 | dev_data->private = ras_features->ctx;
|
| 694 | ret = edac_ecs_get_desc(parent, &ras_attr_groups[attr_gcnt],
|
| 695 | ras_features->ecs_info.num_media_frus);
|
| 696 | if (ret)
|
| 697 | goto data_mem_free;
|
| 698 |
|
| 699 | attr_gcnt += ras_features->ecs_info.num_media_frus;
|
| 700 | break;
|
| 701 | case RAS_FEAT_MEM_REPAIR:
|
| 702 | if (!ras_features->mem_repair_ops ||
|
| 703 | mem_repair_cnt != ras_features->instance) {
|
| 704 | ret = -EINVAL;
|
| 705 | goto data_mem_free;
|
| 706 | }
|
| 707 |
|
| 708 | dev_data = &ctx->mem_repair[mem_repair_cnt];
|
| 709 | dev_data->instance = mem_repair_cnt;
|
| 710 | dev_data->mem_repair_ops = ras_features->mem_repair_ops;
|
| 711 | dev_data->private = ras_features->ctx;
|
| 712 | ret = edac_mem_repair_get_desc(parent, &ras_attr_groups[attr_gcnt],
|
| 713 | ras_features->instance);
|
| 714 | if (ret)
|
| 715 | goto data_mem_free;
|
| 716 |
|
| 717 | mem_repair_cnt++;
|
| 718 | attr_gcnt++;
|
| 719 | break;
|
| 720 | default:
|
| 721 | ret = -EINVAL;
|
| 722 | goto data_mem_free;
|
| 723 | }
|
| 724 | }
|
| 725 |
|
| 726 | ctx->dev.parent = parent;
|
| 727 | ctx->dev.bus = edac_get_sysfs_subsys();
|
| 728 | ctx->dev.type = &edac_dev_type;
|
| 729 | ctx->dev.groups = ras_attr_groups;
|
| 730 | ctx->private = private;
|
| 731 | dev_set_drvdata(&ctx->dev, ctx);
|
| 732 |
|
| 733 | ret = dev_set_name(&ctx->dev, "%s", name);
|
| 734 | if (ret)
|
| 735 | goto data_mem_free;
|
| 736 |
|
| 737 | ret = device_register(&ctx->dev);
|
| 738 | if (ret) {
|
| 739 | put_device(&ctx->dev);
|
| 740 | return ret;
|
| 741 | }
|
| 742 |
|
| 743 | return devm_add_action_or_reset(parent, edac_dev_unreg, &ctx->dev);
|
| 744 |
|
| 745 | data_mem_free:
|
| 746 | kfree(ctx->mem_repair);
|
| 747 | kfree(ctx->scrub);
|
| 748 | groups_free:
|
| 749 | kfree(ras_attr_groups);
|
| 750 | ctx_free:
|
| 751 | kfree(ctx);
|
| 752 | return ret;
|
| 753 | }
|
| 754 | EXPORT_SYMBOL_GPL(edac_dev_register);
|