drivers/edac/edac_device.c

   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 <norsk5@xmission.com>
  10  *
  11  * edac_device API implementation
  12  * 19 Jan 2007
  13  */
  14
  15 #include <linux/module.h>
  16 #include <linux/types.h>
  17 #include <linux/smp.h>
  18 #include <linux/init.h>
  19 #include <linux/sysctl.h>
  20 #include <linux/highmem.h>
  21 #include <linux/timer.h>
  22 #include <linux/slab.h>
  23 #include <linux/spinlock.h>
  24 #include <linux/list.h>
  25 #include <linux/sysdev.h>
  26 #include <linux/ctype.h>
  27 #include <linux/workqueue.h>
  28 #include <asm/uaccess.h>
  29 #include <asm/page.h>
  30
  31 #include "edac_core.h"
  32 #include "edac_module.h"
  33
  34 /* lock to memory controller's control array */
  35 static DECLARE_MUTEX(device_ctls_mutex);
  36 static struct list_head edac_device_list = LIST_HEAD_INIT(edac_device_list);
  37
  38 static inline void lock_device_list(void)
  39 {
  40         down(&device_ctls_mutex);
  41 }
  42
  43 static inline void unlock_device_list(void)
  44 {
  45         up(&device_ctls_mutex);
  46 }
  47
  48 #ifdef CONFIG_EDAC_DEBUG
  49 static void edac_device_dump_device(struct edac_device_ctl_info *edac_dev)
  50 {
  51         debugf3("\tedac_dev = %p dev_idx=%d \n", edac_dev, edac_dev->dev_idx);
  52         debugf4("\tedac_dev->edac_check = %p\n", edac_dev->edac_check);
  53         debugf3("\tdev = %p\n", edac_dev->dev);
  54         debugf3("\tmod_name:ctl_name = %s:%s\n",
  55                 edac_dev->mod_name, edac_dev->ctl_name);
  56         debugf3("\tpvt_info = %p\n\n", edac_dev->pvt_info);
  57 }
  58 #endif                          /* CONFIG_EDAC_DEBUG */
  59
  60 /*
  61  * The alloc() and free() functions for the 'edac_device' control info
  62  * structure. A MC driver will allocate one of these for each edac_device
  63  * it is going to control/register with the EDAC CORE.
  64  */
  65 struct edac_device_ctl_info *edac_device_alloc_ctl_info(
  66         unsigned sz_private,
  67         char *edac_device_name,
  68         unsigned nr_instances,
  69         char *edac_block_name,
  70         unsigned nr_blocks,
  71         unsigned offset_value,
  72         struct edac_attrib_spec
  73         *attrib_spec,
  74         unsigned nr_attribs)
  75 {
  76         struct edac_device_ctl_info *dev_ctl;
  77         struct edac_device_instance *dev_inst, *inst;
  78         struct edac_device_block *dev_blk, *blk_p, *blk;
  79         struct edac_attrib *dev_attrib, *attrib_p, *attrib;
  80         unsigned total_size;
  81         unsigned count;
  82         unsigned instance, block, attr;
  83         void *pvt;
  84
  85         debugf1("%s() instances=%d blocks=%d\n",
  86                 __func__, nr_instances, nr_blocks);
  87
  88         /* Figure out the offsets of the various items from the start of an
  89          * ctl_info structure.  We want the alignment of each item
  90          * to be at least as stringent as what the compiler would
  91          * provide if we could simply hardcode everything into a single struct.
  92          */
  93         dev_ctl = (struct edac_device_ctl_info *)0;
  94
  95         /* Calc the 'end' offset past the ctl_info structure */
  96         dev_inst = (struct edac_device_instance *)
  97             edac_align_ptr(&dev_ctl[1], sizeof(*dev_inst));
  98
  99         /* Calc the 'end' offset past the instance array */
 100         dev_blk = (struct edac_device_block *)
 101             edac_align_ptr(&dev_inst[nr_instances], sizeof(*dev_blk));
 102
 103         /* Calc the 'end' offset past the dev_blk array */
 104         count = nr_instances * nr_blocks;
 105         dev_attrib = (struct edac_attrib *)
 106             edac_align_ptr(&dev_blk[count], sizeof(*dev_attrib));
 107
 108         /* Check for case of NO attributes specified */
 109         if (nr_attribs > 0)
 110                 count *= nr_attribs;
 111
 112         /* Calc the 'end' offset past the attributes array */
 113         pvt = edac_align_ptr(&dev_attrib[count], sz_private);
 114         total_size = ((unsigned long)pvt) + sz_private;
 115
 116         /* Allocate the amount of memory for the set of control structures */
 117         if ((dev_ctl = kmalloc(total_size, GFP_KERNEL)) == NULL)
 118                 return NULL;
 119
 120         /* Adjust pointers so they point within the memory we just allocated
 121          * rather than an imaginary chunk of memory located at address 0.
 122          */
 123         dev_inst = (struct edac_device_instance *)
 124             (((char *)dev_ctl) + ((unsigned long)dev_inst));
 125         dev_blk = (struct edac_device_block *)
 126             (((char *)dev_ctl) + ((unsigned long)dev_blk));
 127         dev_attrib = (struct edac_attrib *)
 128             (((char *)dev_ctl) + ((unsigned long)dev_attrib));
 129         pvt = sz_private ? (((char *)dev_ctl) + ((unsigned long)pvt)) : NULL;
 130
 131         memset(dev_ctl, 0, total_size); /* clear all fields */
 132         dev_ctl->nr_instances = nr_instances;
 133         dev_ctl->instances = dev_inst;
 134         dev_ctl->pvt_info = pvt;
 135
 136         /* Name of this edac device, ensure null terminated */
 137         snprintf(dev_ctl->name, sizeof(dev_ctl->name), "%s", edac_device_name);
 138         dev_ctl->name[sizeof(dev_ctl->name) - 1] = '\0';
 139
 140         /* Initialize every Instance */
 141         for (instance = 0; instance < nr_instances; instance++) {
 142                 inst = &dev_inst[instance];
 143                 inst->ctl = dev_ctl;
 144                 inst->nr_blocks = nr_blocks;
 145                 blk_p = &dev_blk[instance * nr_blocks];
 146                 inst->blocks = blk_p;
 147
 148                 /* name of this instance */
 149                 snprintf(inst->name, sizeof(inst->name),
 150                          "%s%u", edac_device_name, instance);
 151                 inst->name[sizeof(inst->name) - 1] = '\0';
 152
 153                 /* Initialize every block in each instance */
 154                 for (block = 0; block < nr_blocks; block++) {
 155                         blk = &blk_p[block];
 156                         blk->instance = inst;
 157                         blk->nr_attribs = nr_attribs;
 158                         attrib_p = &dev_attrib[block * nr_attribs];
 159                         blk->attribs = attrib_p;
 160                         snprintf(blk->name, sizeof(blk->name),
 161                                  "%s%d", edac_block_name, block + 1);
 162                         blk->name[sizeof(blk->name) - 1] = '\0';
 163
 164                         debugf1("%s() instance=%d block=%d name=%s\n",
 165                                 __func__, instance, block, blk->name);
 166
 167                         if (attrib_spec != NULL) {
 168                                 /* when there is an attrib_spec passed int then
 169                                  * Initialize every attrib of each block
 170                                  */
 171                                 for (attr = 0; attr < nr_attribs; attr++) {
 172                                         attrib = &attrib_p[attr];
 173                                         attrib->block = blk;
 174
 175                                         /* Link each attribute to the caller's
 176                                          * spec entry, for name and type
 177                                          */
 178                                         attrib->spec = &attrib_spec[attr];
 179                                 }
 180                         }
 181                 }
 182         }
 183
 184         /* Mark this instance as merely ALLOCATED */
 185         dev_ctl->op_state = OP_ALLOC;
 186
 187         return dev_ctl;
 188 }
 189
 190 EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
 191
 192 /*
 193  * edac_device_free_ctl_info()
 194  *      frees the memory allocated by the edac_device_alloc_ctl_info()
 195  *      function
 196  */
 197 void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)
 198 {
 199         kfree(ctl_info);
 200 }
 201
 202 EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
 203
 204 /*
 205  * find_edac_device_by_dev
 206  *      scans the edac_device list for a specific 'struct device *'
 207  */
 208 static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
 209 {
 210         struct edac_device_ctl_info *edac_dev;
 211         struct list_head *item;
 212
 213         debugf3("%s()\n", __func__);
 214
 215         list_for_each(item, &edac_device_list) {
 216                 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
 217
 218                 if (edac_dev->dev == dev)
 219                         return edac_dev;
 220         }
 221
 222         return NULL;
 223 }
 224
 225 /*
 226  * add_edac_dev_to_global_list
 227  *      Before calling this function, caller must
 228  *      assign a unique value to edac_dev->dev_idx.
 229  *      Return:
 230  *              0 on success
 231  *              1 on failure.
 232  */
 233 static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
 234 {
 235         struct list_head *item, *insert_before;
 236         struct edac_device_ctl_info *rover;
 237
 238         insert_before = &edac_device_list;
 239
 240         /* Determine if already on the list */
 241         if (unlikely((rover = find_edac_device_by_dev(edac_dev->dev)) != NULL))
 242                 goto fail0;
 243
 244         /* Insert in ascending order by 'dev_idx', so find position */
 245         list_for_each(item, &edac_device_list) {
 246                 rover = list_entry(item, struct edac_device_ctl_info, link);
 247
 248                 if (rover->dev_idx >= edac_dev->dev_idx) {
 249                         if (unlikely(rover->dev_idx == edac_dev->dev_idx))
 250                                 goto fail1;
 251
 252                         insert_before = item;
 253                         break;
 254                 }
 255         }
 256
 257         list_add_tail_rcu(&edac_dev->link, insert_before);
 258         return 0;
 259
 260       fail0:
 261         edac_printk(KERN_WARNING, EDAC_MC,
 262                     "%s (%s) %s %s already assigned %d\n",
 263                     rover->dev->bus_id, dev_name(rover),
 264                     rover->mod_name, rover->ctl_name, rover->dev_idx);
 265         return 1;
 266
 267       fail1:
 268         edac_printk(KERN_WARNING, EDAC_MC,
 269                     "bug in low-level driver: attempt to assign\n"
 270                     "    duplicate dev_idx %d in %s()\n", rover->dev_idx,
 271                     __func__);
 272         return 1;
 273 }
 274
 275 /*
 276  * complete_edac_device_list_del
 277  */
 278 static void complete_edac_device_list_del(struct rcu_head *head)
 279 {
 280         struct edac_device_ctl_info *edac_dev;
 281
 282         edac_dev = container_of(head, struct edac_device_ctl_info, rcu);
 283         INIT_LIST_HEAD(&edac_dev->link);
 284         complete(&edac_dev->complete);
 285 }
 286
 287 /*
 288  * del_edac_device_from_global_list
 289  */
 290 static void del_edac_device_from_global_list(struct edac_device_ctl_info
 291                                              *edac_device)
 292 {
 293         list_del_rcu(&edac_device->link);
 294         init_completion(&edac_device->complete);
 295         call_rcu(&edac_device->rcu, complete_edac_device_list_del);
 296         wait_for_completion(&edac_device->complete);
 297 }
 298
 299 /**
 300  * edac_device_find
 301  *      Search for a edac_device_ctl_info structure whose index is 'idx'.
 302  *
 303  * If found, return a pointer to the structure.
 304  * Else return NULL.
 305  *
 306  * Caller must hold device_ctls_mutex.
 307  */
 308 struct edac_device_ctl_info *edac_device_find(int idx)
 309 {
 310         struct list_head *item;
 311         struct edac_device_ctl_info *edac_dev;
 312
 313         /* Iterate over list, looking for exact match of ID */
 314         list_for_each(item, &edac_device_list) {
 315                 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
 316
 317                 if (edac_dev->dev_idx >= idx) {
 318                         if (edac_dev->dev_idx == idx)
 319                                 return edac_dev;
 320
 321                         /* not on list, so terminate early */
 322                         break;
 323                 }
 324         }
 325
 326         return NULL;
 327 }
 328
 329 EXPORT_SYMBOL(edac_device_find);
 330
 331 /*
 332  * edac_device_workq_function
 333  *      performs the operation scheduled by a workq request
 334  */
 335 static void edac_device_workq_function(struct work_struct *work_req)
 336 {
 337         struct delayed_work *d_work = (struct delayed_work *)work_req;
 338         struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);
 339
 340         //debugf0("%s() here and running\n", __func__);
 341         lock_device_list();
 342
 343         /* Only poll controllers that are running polled and have a check */
 344         if ((edac_dev->op_state == OP_RUNNING_POLL) &&
 345             (edac_dev->edac_check != NULL)) {
 346                 edac_dev->edac_check(edac_dev);
 347         }
 348
 349         unlock_device_list();
 350
 351         /* Reschedule */
 352         queue_delayed_work(edac_workqueue, &edac_dev->work, edac_dev->delay);
 353 }
 354
 355 /*
 356  * edac_device_workq_setup
 357  *      initialize a workq item for this edac_device instance
 358  *      passing in the new delay period in msec
 359  */
 360 void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
 361                              unsigned msec)
 362 {
 363         debugf0("%s()\n", __func__);
 364
 365         edac_dev->poll_msec = msec;
 366         edac_calc_delay(edac_dev);      /* Calc delay jiffies */
 367
 368         INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
 369         queue_delayed_work(edac_workqueue, &edac_dev->work, edac_dev->delay);
 370 }
 371
 372 /*
 373  * edac_device_workq_teardown
 374  *      stop the workq processing on this edac_dev
 375  */
 376 void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)
 377 {
 378         int status;
 379
 380         status = cancel_delayed_work(&edac_dev->work);
 381         if (status == 0) {
 382                 /* workq instance might be running, wait for it */
 383                 flush_workqueue(edac_workqueue);
 384         }
 385 }
 386
 387 /*
 388  * edac_device_reset_delay_period
 389  */
 390
 391 void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,
 392                                     unsigned long value)
 393 {
 394         lock_device_list();
 395
 396         /* cancel the current workq request */
 397         edac_device_workq_teardown(edac_dev);
 398
 399         /* restart the workq request, with new delay value */
 400         edac_device_workq_setup(edac_dev, value);
 401
 402         unlock_device_list();
 403 }
 404
 405 /**
 406  * edac_device_add_device: Insert the 'edac_dev' structure into the
 407  * edac_device global list and create sysfs entries associated with
 408  * edac_device structure.
 409  * @edac_device: pointer to the edac_device structure to be added to the list
 410  * @edac_idx: A unique numeric identifier to be assigned to the
 411  * 'edac_device' structure.
 412  *
 413  * Return:
 414  *      0       Success
 415  *      !0      Failure
 416  */
 417 int edac_device_add_device(struct edac_device_ctl_info *edac_dev, int edac_idx)
 418 {
 419         debugf0("%s()\n", __func__);
 420
 421         edac_dev->dev_idx = edac_idx;
 422 #ifdef CONFIG_EDAC_DEBUG
 423         if (edac_debug_level >= 3)
 424                 edac_device_dump_device(edac_dev);
 425 #endif
 426         lock_device_list();
 427
 428         if (add_edac_dev_to_global_list(edac_dev))
 429                 goto fail0;
 430
 431         /* set load time so that error rate can be tracked */
 432         edac_dev->start_time = jiffies;
 433
 434         /* create this instance's sysfs entries */
 435         if (edac_device_create_sysfs(edac_dev)) {
 436                 edac_device_printk(edac_dev, KERN_WARNING,
 437                                    "failed to create sysfs device\n");
 438                 goto fail1;
 439         }
 440
 441         /* If there IS a check routine, then we are running POLLED */
 442         if (edac_dev->edac_check != NULL) {
 443                 /* This instance is NOW RUNNING */
 444                 edac_dev->op_state = OP_RUNNING_POLL;
 445
 446                 /*
 447                  * enable workq processing on this instance,
 448                  * default = 1000 msec
 449                  */
 450                 edac_device_workq_setup(edac_dev, 1000);
 451         } else {
 452                 edac_dev->op_state = OP_RUNNING_INTERRUPT;
 453         }
 454
 455         /* Report action taken */
 456         edac_device_printk(edac_dev, KERN_INFO,
 457                            "Giving out device to module '%s' controller '%s': DEV '%s' (%s)\n",
 458                            edac_dev->mod_name,
 459                            edac_dev->ctl_name,
 460                            dev_name(edac_dev),
 461                            edac_op_state_toString(edac_dev->op_state)
 462             );
 463
 464         unlock_device_list();
 465         return 0;
 466
 467       fail1:
 468         /* Some error, so remove the entry from the lsit */
 469         del_edac_device_from_global_list(edac_dev);
 470
 471       fail0:
 472         unlock_device_list();
 473         return 1;
 474 }
 475
 476 EXPORT_SYMBOL_GPL(edac_device_add_device);
 477
 478 /**
 479  * edac_device_del_device:
 480  *      Remove sysfs entries for specified edac_device structure and
 481  *      then remove edac_device structure from global list
 482  *
 483  * @pdev:
 484  *      Pointer to 'struct device' representing edac_device
 485  *      structure to remove.
 486  *
 487  * Return:
 488  *      Pointer to removed edac_device structure,
 489  *      OR NULL if device not found.
 490  */
 491 struct edac_device_ctl_info *edac_device_del_device(struct device *dev)
 492 {
 493         struct edac_device_ctl_info *edac_dev;
 494
 495         debugf0("MC: %s()\n", __func__);
 496
 497         lock_device_list();
 498
 499         if ((edac_dev = find_edac_device_by_dev(dev)) == NULL) {
 500                 unlock_device_list();
 501                 return NULL;
 502         }
 503
 504         /* mark this instance as OFFLINE */
 505         edac_dev->op_state = OP_OFFLINE;
 506
 507         /* clear workq processing on this instance */
 508         edac_device_workq_teardown(edac_dev);
 509
 510         /* Tear down the sysfs entries for this instance */
 511         edac_device_remove_sysfs(edac_dev);
 512
 513         /* deregister from global list */
 514         del_edac_device_from_global_list(edac_dev);
 515
 516         unlock_device_list();
 517
 518         edac_printk(KERN_INFO, EDAC_MC,
 519                     "Removed device %d for %s %s: DEV %s\n",
 520                     edac_dev->dev_idx,
 521                     edac_dev->mod_name, edac_dev->ctl_name, dev_name(edac_dev));
 522
 523         return edac_dev;
 524 }
 525
 526 EXPORT_SYMBOL_GPL(edac_device_del_device);
 527
 528 static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
 529 {
 530         return edac_dev->log_ce;
 531 }
 532
 533 static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
 534 {
 535         return edac_dev->log_ue;
 536 }
 537
 538 static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
 539                                               *edac_dev)
 540 {
 541         return edac_dev->panic_on_ue;
 542 }
 543
 544 /*
 545  * edac_device_handle_ce
 546  *      perform a common output and handling of an 'edac_dev' CE event
 547  */
 548 void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
 549                            int inst_nr, int block_nr, const char *msg)
 550 {
 551         struct edac_device_instance *instance;
 552         struct edac_device_block *block = NULL;
 553
 554         if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
 555                 edac_device_printk(edac_dev, KERN_ERR,
 556                                    "INTERNAL ERROR: 'instance' out of range "
 557                                    "(%d >= %d)\n", inst_nr,
 558                                    edac_dev->nr_instances);
 559                 return;
 560         }
 561
 562         instance = edac_dev->instances + inst_nr;
 563
 564         if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
 565                 edac_device_printk(edac_dev, KERN_ERR,
 566                                    "INTERNAL ERROR: instance %d 'block' out of range "
 567                                    "(%d >= %d)\n", inst_nr, block_nr,
 568                                    instance->nr_blocks);
 569                 return;
 570         }
 571
 572         if (instance->nr_blocks > 0) {
 573                 block = instance->blocks + block_nr;
 574                 block->counters.ce_count++;
 575         }
 576
 577         /* Propogate the count up the 'totals' tree */
 578         instance->counters.ce_count++;
 579         edac_dev->counters.ce_count++;
 580
 581         if (edac_device_get_log_ce(edac_dev))
 582                 edac_device_printk(edac_dev, KERN_WARNING,
 583                                    "CE ctl: %s, instance: %s, block: %s: %s\n",
 584                                    edac_dev->ctl_name, instance->name,
 585                                    block ? block->name : "N/A", msg);
 586 }
 587
 588 EXPORT_SYMBOL_GPL(edac_device_handle_ce);
 589
 590 /*
 591  * edac_device_handle_ue
 592  *      perform a common output and handling of an 'edac_dev' UE event
 593  */
 594 void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
 595                            int inst_nr, int block_nr, const char *msg)
 596 {
 597         struct edac_device_instance *instance;
 598         struct edac_device_block *block = NULL;
 599
 600         if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
 601                 edac_device_printk(edac_dev, KERN_ERR,
 602                                    "INTERNAL ERROR: 'instance' out of range "
 603                                    "(%d >= %d)\n", inst_nr,
 604                                    edac_dev->nr_instances);
 605                 return;
 606         }
 607
 608         instance = edac_dev->instances + inst_nr;
 609
 610         if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
 611                 edac_device_printk(edac_dev, KERN_ERR,
 612                                    "INTERNAL ERROR: instance %d 'block' out of range "
 613                                    "(%d >= %d)\n", inst_nr, block_nr,
 614                                    instance->nr_blocks);
 615                 return;
 616         }
 617
 618         if (instance->nr_blocks > 0) {
 619                 block = instance->blocks + block_nr;
 620                 block->counters.ue_count++;
 621         }
 622
 623         /* Propogate the count up the 'totals' tree */
 624         instance->counters.ue_count++;
 625         edac_dev->counters.ue_count++;
 626
 627         if (edac_device_get_log_ue(edac_dev))
 628                 edac_device_printk(edac_dev, KERN_EMERG,
 629                                    "UE ctl: %s, instance: %s, block: %s: %s\n",
 630                                    edac_dev->ctl_name, instance->name,
 631                                    block ? block->name : "N/A", msg);
 632
 633         if (edac_device_get_panic_on_ue(edac_dev))
 634                 panic("EDAC %s: UE instance: %s, block %s: %s\n",
 635                       edac_dev->ctl_name, instance->name,
 636                       block ? block->name : "N/A", msg);
 637 }
 638
 639 EXPORT_SYMBOL_GPL(edac_device_handle_ue);