573 lines · c
1 2/*3 * edac_device.c4 * (C) 2007 www.douglaskthompson.com5 *6 * This file may be distributed under the terms of the7 * GNU General Public License.8 *9 * Written by Doug Thompson <norsk5@xmission.com>10 *11 * edac_device API implementation12 * 19 Jan 200713 */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/* lock for the list: 'edac_device_list', manipulation of this list32 * is protected by the 'device_ctls_mutex' lock33 */34static DEFINE_MUTEX(device_ctls_mutex);35static LIST_HEAD(edac_device_list);36 37/* Default workqueue processing interval on this instance, in msecs */38#define DEFAULT_POLL_INTERVAL 100039 40#ifdef CONFIG_EDAC_DEBUG41static 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 /* CONFIG_EDAC_DEBUG */52 53/*54 * @off_val: zero, 1, or other based offset55 */56struct edac_device_ctl_info *57edac_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(sizeof(struct edac_device_ctl_info), GFP_KERNEL);71 if (!dev_ctl)72 return NULL;73 74 dev_inst = kcalloc(nr_instances, sizeof(struct edac_device_instance), GFP_KERNEL);75 if (!dev_inst)76 goto free;77 78 dev_ctl->instances = dev_inst;79 80 dev_blk = kcalloc(nr_instances * nr_blocks, sizeof(struct edac_device_block), GFP_KERNEL);81 if (!dev_blk)82 goto free;83 84 dev_ctl->blocks = dev_blk;85 86 if (pvt_sz) {87 pvt = kzalloc(pvt_sz, GFP_KERNEL);88 if (!pvt)89 goto free;90 91 dev_ctl->pvt_info = pvt;92 }93 94 dev_ctl->dev_idx = device_index;95 dev_ctl->nr_instances = nr_instances;96 97 /* Default logging of CEs and UEs */98 dev_ctl->log_ce = 1;99 dev_ctl->log_ue = 1;100 101 /* Name of this edac device */102 snprintf(dev_ctl->name, sizeof(dev_ctl->name),"%s", dev_name);103 104 /* Initialize every Instance */105 for (instance = 0; instance < nr_instances; instance++) {106 inst = &dev_inst[instance];107 inst->ctl = dev_ctl;108 inst->nr_blocks = nr_blocks;109 blk_p = &dev_blk[instance * nr_blocks];110 inst->blocks = blk_p;111 112 /* name of this instance */113 snprintf(inst->name, sizeof(inst->name), "%s%u", dev_name, instance);114 115 /* Initialize every block in each instance */116 for (block = 0; block < nr_blocks; block++) {117 blk = &blk_p[block];118 blk->instance = inst;119 snprintf(blk->name, sizeof(blk->name),120 "%s%d", blk_name, block + off_val);121 122 edac_dbg(4, "instance=%d inst_p=%p block=#%d block_p=%p name='%s'\n",123 instance, inst, block, blk, blk->name);124 }125 }126 127 /* Mark this instance as merely ALLOCATED */128 dev_ctl->op_state = OP_ALLOC;129 130 /*131 * Initialize the 'root' kobj for the edac_device controller132 */133 err = edac_device_register_sysfs_main_kobj(dev_ctl);134 if (err)135 goto free;136 137 /* at this point, the root kobj is valid, and in order to138 * 'free' the object, then the function:139 * edac_device_unregister_sysfs_main_kobj() must be called140 * which will perform kobj unregistration and the actual free141 * will occur during the kobject callback operation142 */143 144 return dev_ctl;145 146free:147 __edac_device_free_ctl_info(dev_ctl);148 149 return NULL;150}151EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);152 153void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)154{155 edac_device_unregister_sysfs_main_kobj(ctl_info);156}157EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);158 159/*160 * find_edac_device_by_dev161 * scans the edac_device list for a specific 'struct device *'162 *163 * lock to be held prior to call: device_ctls_mutex164 *165 * Return:166 * pointer to control structure managing 'dev'167 * NULL if not found on list168 */169static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)170{171 struct edac_device_ctl_info *edac_dev;172 struct list_head *item;173 174 edac_dbg(0, "\n");175 176 list_for_each(item, &edac_device_list) {177 edac_dev = list_entry(item, struct edac_device_ctl_info, link);178 179 if (edac_dev->dev == dev)180 return edac_dev;181 }182 183 return NULL;184}185 186/*187 * add_edac_dev_to_global_list188 * Before calling this function, caller must189 * assign a unique value to edac_dev->dev_idx.190 *191 * lock to be held prior to call: device_ctls_mutex192 *193 * Return:194 * 0 on success195 * 1 on failure.196 */197static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)198{199 struct list_head *item, *insert_before;200 struct edac_device_ctl_info *rover;201 202 insert_before = &edac_device_list;203 204 /* Determine if already on the list */205 rover = find_edac_device_by_dev(edac_dev->dev);206 if (unlikely(rover != NULL))207 goto fail0;208 209 /* Insert in ascending order by 'dev_idx', so find position */210 list_for_each(item, &edac_device_list) {211 rover = list_entry(item, struct edac_device_ctl_info, link);212 213 if (rover->dev_idx >= edac_dev->dev_idx) {214 if (unlikely(rover->dev_idx == edac_dev->dev_idx))215 goto fail1;216 217 insert_before = item;218 break;219 }220 }221 222 list_add_tail_rcu(&edac_dev->link, insert_before);223 return 0;224 225fail0:226 edac_printk(KERN_WARNING, EDAC_MC,227 "%s (%s) %s %s already assigned %d\n",228 dev_name(rover->dev), edac_dev_name(rover),229 rover->mod_name, rover->ctl_name, rover->dev_idx);230 return 1;231 232fail1:233 edac_printk(KERN_WARNING, EDAC_MC,234 "bug in low-level driver: attempt to assign\n"235 " duplicate dev_idx %d in %s()\n", rover->dev_idx,236 __func__);237 return 1;238}239 240/*241 * del_edac_device_from_global_list242 */243static void del_edac_device_from_global_list(struct edac_device_ctl_info244 *edac_device)245{246 list_del_rcu(&edac_device->link);247 248 /* these are for safe removal of devices from global list while249 * NMI handlers may be traversing list250 */251 synchronize_rcu();252 INIT_LIST_HEAD(&edac_device->link);253}254 255/*256 * edac_device_workq_function257 * performs the operation scheduled by a workq request258 *259 * this workq is embedded within an edac_device_ctl_info260 * structure, that needs to be polled for possible error events.261 *262 * This operation is to acquire the list mutex lock263 * (thus preventing insertation or deletion)264 * and then call the device's poll function IFF this device is265 * running polled and there is a poll function defined.266 */267static void edac_device_workq_function(struct work_struct *work_req)268{269 struct delayed_work *d_work = to_delayed_work(work_req);270 struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);271 272 mutex_lock(&device_ctls_mutex);273 274 /* If we are being removed, bail out immediately */275 if (edac_dev->op_state == OP_OFFLINE) {276 mutex_unlock(&device_ctls_mutex);277 return;278 }279 280 /* Only poll controllers that are running polled and have a check */281 if ((edac_dev->op_state == OP_RUNNING_POLL) &&282 (edac_dev->edac_check != NULL)) {283 edac_dev->edac_check(edac_dev);284 }285 286 mutex_unlock(&device_ctls_mutex);287 288 /* Reschedule the workq for the next time period to start again289 * if the number of msec is for 1 sec, then adjust to the next290 * whole one second to save timers firing all over the period291 * between integral seconds292 */293 if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)294 edac_queue_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));295 else296 edac_queue_work(&edac_dev->work, edac_dev->delay);297}298 299/*300 * edac_device_workq_setup301 * initialize a workq item for this edac_device instance302 * passing in the new delay period in msec303 */304static void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,305 unsigned msec)306{307 edac_dbg(0, "\n");308 309 /* take the arg 'msec' and set it into the control structure310 * to used in the time period calculation311 * then calc the number of jiffies that represents312 */313 edac_dev->poll_msec = msec;314 edac_dev->delay = msecs_to_jiffies(msec);315 316 INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);317 318 /* optimize here for the 1 second case, which will be normal value, to319 * fire ON the 1 second time event. This helps reduce all sorts of320 * timers firing on sub-second basis, while they are happy321 * to fire together on the 1 second exactly322 */323 if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)324 edac_queue_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));325 else326 edac_queue_work(&edac_dev->work, edac_dev->delay);327}328 329/*330 * edac_device_workq_teardown331 * stop the workq processing on this edac_dev332 */333static void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)334{335 if (!edac_dev->edac_check)336 return;337 338 edac_dev->op_state = OP_OFFLINE;339 340 edac_stop_work(&edac_dev->work);341}342 343/*344 * edac_device_reset_delay_period345 *346 * need to stop any outstanding workq queued up at this time347 * because we will be resetting the sleep time.348 * Then restart the workq on the new delay349 */350void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,351 unsigned long msec)352{353 edac_dev->poll_msec = msec;354 edac_dev->delay = msecs_to_jiffies(msec);355 356 /* See comment in edac_device_workq_setup() above */357 if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)358 edac_mod_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));359 else360 edac_mod_work(&edac_dev->work, edac_dev->delay);361}362 363int edac_device_alloc_index(void)364{365 static atomic_t device_indexes = ATOMIC_INIT(0);366 367 return atomic_inc_return(&device_indexes) - 1;368}369EXPORT_SYMBOL_GPL(edac_device_alloc_index);370 371int edac_device_add_device(struct edac_device_ctl_info *edac_dev)372{373 edac_dbg(0, "\n");374 375#ifdef CONFIG_EDAC_DEBUG376 if (edac_debug_level >= 3)377 edac_device_dump_device(edac_dev);378#endif379 mutex_lock(&device_ctls_mutex);380 381 if (add_edac_dev_to_global_list(edac_dev))382 goto fail0;383 384 /* set load time so that error rate can be tracked */385 edac_dev->start_time = jiffies;386 387 /* create this instance's sysfs entries */388 if (edac_device_create_sysfs(edac_dev)) {389 edac_device_printk(edac_dev, KERN_WARNING,390 "failed to create sysfs device\n");391 goto fail1;392 }393 394 /* If there IS a check routine, then we are running POLLED */395 if (edac_dev->edac_check != NULL) {396 /* This instance is NOW RUNNING */397 edac_dev->op_state = OP_RUNNING_POLL;398 399 edac_device_workq_setup(edac_dev, edac_dev->poll_msec ?: DEFAULT_POLL_INTERVAL);400 } else {401 edac_dev->op_state = OP_RUNNING_INTERRUPT;402 }403 404 /* Report action taken */405 edac_device_printk(edac_dev, KERN_INFO,406 "Giving out device to module %s controller %s: DEV %s (%s)\n",407 edac_dev->mod_name, edac_dev->ctl_name, edac_dev->dev_name,408 edac_op_state_to_string(edac_dev->op_state));409 410 mutex_unlock(&device_ctls_mutex);411 return 0;412 413fail1:414 /* Some error, so remove the entry from the lsit */415 del_edac_device_from_global_list(edac_dev);416 417fail0:418 mutex_unlock(&device_ctls_mutex);419 return 1;420}421EXPORT_SYMBOL_GPL(edac_device_add_device);422 423struct edac_device_ctl_info *edac_device_del_device(struct device *dev)424{425 struct edac_device_ctl_info *edac_dev;426 427 edac_dbg(0, "\n");428 429 mutex_lock(&device_ctls_mutex);430 431 /* Find the structure on the list, if not there, then leave */432 edac_dev = find_edac_device_by_dev(dev);433 if (edac_dev == NULL) {434 mutex_unlock(&device_ctls_mutex);435 return NULL;436 }437 438 /* mark this instance as OFFLINE */439 edac_dev->op_state = OP_OFFLINE;440 441 /* deregister from global list */442 del_edac_device_from_global_list(edac_dev);443 444 mutex_unlock(&device_ctls_mutex);445 446 /* clear workq processing on this instance */447 edac_device_workq_teardown(edac_dev);448 449 /* Tear down the sysfs entries for this instance */450 edac_device_remove_sysfs(edac_dev);451 452 edac_printk(KERN_INFO, EDAC_MC,453 "Removed device %d for %s %s: DEV %s\n",454 edac_dev->dev_idx,455 edac_dev->mod_name, edac_dev->ctl_name, edac_dev_name(edac_dev));456 457 return edac_dev;458}459EXPORT_SYMBOL_GPL(edac_device_del_device);460 461static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)462{463 return edac_dev->log_ce;464}465 466static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)467{468 return edac_dev->log_ue;469}470 471static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info472 *edac_dev)473{474 return edac_dev->panic_on_ue;475}476 477void edac_device_handle_ce_count(struct edac_device_ctl_info *edac_dev,478 unsigned int count, int inst_nr, int block_nr,479 const char *msg)480{481 struct edac_device_instance *instance;482 struct edac_device_block *block = NULL;483 484 if (!count)485 return;486 487 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {488 edac_device_printk(edac_dev, KERN_ERR,489 "INTERNAL ERROR: 'instance' out of range "490 "(%d >= %d)\n", inst_nr,491 edac_dev->nr_instances);492 return;493 }494 495 instance = edac_dev->instances + inst_nr;496 497 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {498 edac_device_printk(edac_dev, KERN_ERR,499 "INTERNAL ERROR: instance %d 'block' "500 "out of range (%d >= %d)\n",501 inst_nr, block_nr,502 instance->nr_blocks);503 return;504 }505 506 if (instance->nr_blocks > 0) {507 block = instance->blocks + block_nr;508 block->counters.ce_count += count;509 }510 511 /* Propagate the count up the 'totals' tree */512 instance->counters.ce_count += count;513 edac_dev->counters.ce_count += count;514 515 if (edac_device_get_log_ce(edac_dev))516 edac_device_printk(edac_dev, KERN_WARNING,517 "CE: %s instance: %s block: %s count: %d '%s'\n",518 edac_dev->ctl_name, instance->name,519 block ? block->name : "N/A", count, msg);520}521EXPORT_SYMBOL_GPL(edac_device_handle_ce_count);522 523void edac_device_handle_ue_count(struct edac_device_ctl_info *edac_dev,524 unsigned int count, int inst_nr, int block_nr,525 const char *msg)526{527 struct edac_device_instance *instance;528 struct edac_device_block *block = NULL;529 530 if (!count)531 return;532 533 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {534 edac_device_printk(edac_dev, KERN_ERR,535 "INTERNAL ERROR: 'instance' out of range "536 "(%d >= %d)\n", inst_nr,537 edac_dev->nr_instances);538 return;539 }540 541 instance = edac_dev->instances + inst_nr;542 543 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {544 edac_device_printk(edac_dev, KERN_ERR,545 "INTERNAL ERROR: instance %d 'block' "546 "out of range (%d >= %d)\n",547 inst_nr, block_nr,548 instance->nr_blocks);549 return;550 }551 552 if (instance->nr_blocks > 0) {553 block = instance->blocks + block_nr;554 block->counters.ue_count += count;555 }556 557 /* Propagate the count up the 'totals' tree */558 instance->counters.ue_count += count;559 edac_dev->counters.ue_count += count;560 561 if (edac_device_get_log_ue(edac_dev))562 edac_device_printk(edac_dev, KERN_EMERG,563 "UE: %s instance: %s block: %s count: %d '%s'\n",564 edac_dev->ctl_name, instance->name,565 block ? block->name : "N/A", count, msg);566 567 if (edac_device_get_panic_on_ue(edac_dev))568 panic("EDAC %s: UE instance: %s block %s count: %d '%s'\n",569 edac_dev->ctl_name, instance->name,570 block ? block->name : "N/A", count, msg);571}572EXPORT_SYMBOL_GPL(edac_device_handle_ue_count);573