Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
edac_device.c
Go to the documentation of this file.
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 <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/jiffies.h>
24 #include <linux/spinlock.h>
25 #include <linux/list.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 for the list: 'edac_device_list', manipulation of this list
35  * is protected by the 'device_ctls_mutex' lock
36  */
37 static DEFINE_MUTEX(device_ctls_mutex);
38 static LIST_HEAD(edac_device_list);
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 /* CONFIG_EDAC_DEBUG */
52 
53 
54 /*
55  * edac_device_alloc_ctl_info()
56  * Allocate a new edac device control info structure
57  *
58  * The control structure is allocated in complete chunk
59  * from the OS. It is in turn sub allocated to the
60  * various objects that compose the structure
61  *
62  * The structure has a 'nr_instance' array within itself.
63  * Each instance represents a major component
64  * Example: L1 cache and L2 cache are 2 instance components
65  *
66  * Within each instance is an array of 'nr_blocks' blockoffsets
67  */
69  unsigned sz_private,
70  char *edac_device_name, unsigned nr_instances,
71  char *edac_block_name, unsigned nr_blocks,
72  unsigned offset_value, /* zero, 1, or other based offset */
73  struct edac_dev_sysfs_block_attribute *attrib_spec, unsigned nr_attrib,
74  int device_index)
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_dev_sysfs_block_attribute *dev_attrib, *attrib_p, *attrib;
80  unsigned total_size;
81  unsigned count;
82  unsigned instance, block, attr;
83  void *pvt, *p;
84  int err;
85 
86  edac_dbg(4, "instances=%d blocks=%d\n", nr_instances, nr_blocks);
87 
88  /* Calculate the size of memory we need to allocate AND
89  * determine the offsets of the various item arrays
90  * (instance,block,attrib) from the start of an allocated structure.
91  * We want the alignment of each item (instance,block,attrib)
92  * to be at least as stringent as what the compiler would
93  * provide if we could simply hardcode everything into a single struct.
94  */
95  p = NULL;
96  dev_ctl = edac_align_ptr(&p, sizeof(*dev_ctl), 1);
97 
98  /* Calc the 'end' offset past end of ONE ctl_info structure
99  * which will become the start of the 'instance' array
100  */
101  dev_inst = edac_align_ptr(&p, sizeof(*dev_inst), nr_instances);
102 
103  /* Calc the 'end' offset past the instance array within the ctl_info
104  * which will become the start of the block array
105  */
106  count = nr_instances * nr_blocks;
107  dev_blk = edac_align_ptr(&p, sizeof(*dev_blk), count);
108 
109  /* Calc the 'end' offset past the dev_blk array
110  * which will become the start of the attrib array, if any.
111  */
112  /* calc how many nr_attrib we need */
113  if (nr_attrib > 0)
114  count *= nr_attrib;
115  dev_attrib = edac_align_ptr(&p, sizeof(*dev_attrib), count);
116 
117  /* Calc the 'end' offset past the attributes array */
118  pvt = edac_align_ptr(&p, sz_private, 1);
119 
120  /* 'pvt' now points to where the private data area is.
121  * At this point 'pvt' (like dev_inst,dev_blk and dev_attrib)
122  * is baselined at ZERO
123  */
124  total_size = ((unsigned long)pvt) + sz_private;
125 
126  /* Allocate the amount of memory for the set of control structures */
127  dev_ctl = kzalloc(total_size, GFP_KERNEL);
128  if (dev_ctl == NULL)
129  return NULL;
130 
131  /* Adjust pointers so they point within the actual memory we
132  * just allocated rather than an imaginary chunk of memory
133  * located at address 0.
134  * 'dev_ctl' points to REAL memory, while the others are
135  * ZERO based and thus need to be adjusted to point within
136  * the allocated memory.
137  */
138  dev_inst = (struct edac_device_instance *)
139  (((char *)dev_ctl) + ((unsigned long)dev_inst));
140  dev_blk = (struct edac_device_block *)
141  (((char *)dev_ctl) + ((unsigned long)dev_blk));
142  dev_attrib = (struct edac_dev_sysfs_block_attribute *)
143  (((char *)dev_ctl) + ((unsigned long)dev_attrib));
144  pvt = sz_private ? (((char *)dev_ctl) + ((unsigned long)pvt)) : NULL;
145 
146  /* Begin storing the information into the control info structure */
147  dev_ctl->dev_idx = device_index;
148  dev_ctl->nr_instances = nr_instances;
149  dev_ctl->instances = dev_inst;
150  dev_ctl->pvt_info = pvt;
151 
152  /* Default logging of CEs and UEs */
153  dev_ctl->log_ce = 1;
154  dev_ctl->log_ue = 1;
155 
156  /* Name of this edac device */
157  snprintf(dev_ctl->name,sizeof(dev_ctl->name),"%s",edac_device_name);
158 
159  edac_dbg(4, "edac_dev=%p next after end=%p\n",
160  dev_ctl, pvt + sz_private);
161 
162  /* Initialize every Instance */
163  for (instance = 0; instance < nr_instances; instance++) {
164  inst = &dev_inst[instance];
165  inst->ctl = dev_ctl;
166  inst->nr_blocks = nr_blocks;
167  blk_p = &dev_blk[instance * nr_blocks];
168  inst->blocks = blk_p;
169 
170  /* name of this instance */
171  snprintf(inst->name, sizeof(inst->name),
172  "%s%u", edac_device_name, instance);
173 
174  /* Initialize every block in each instance */
175  for (block = 0; block < nr_blocks; block++) {
176  blk = &blk_p[block];
177  blk->instance = inst;
178  snprintf(blk->name, sizeof(blk->name),
179  "%s%d", edac_block_name, block+offset_value);
180 
181  edac_dbg(4, "instance=%d inst_p=%p block=#%d block_p=%p name='%s'\n",
182  instance, inst, block, blk, blk->name);
183 
184  /* if there are NO attributes OR no attribute pointer
185  * then continue on to next block iteration
186  */
187  if ((nr_attrib == 0) || (attrib_spec == NULL))
188  continue;
189 
190  /* setup the attribute array for this block */
191  blk->nr_attribs = nr_attrib;
192  attrib_p = &dev_attrib[block*nr_instances*nr_attrib];
193  blk->block_attributes = attrib_p;
194 
195  edac_dbg(4, "THIS BLOCK_ATTRIB=%p\n",
196  blk->block_attributes);
197 
198  /* Initialize every user specified attribute in this
199  * block with the data the caller passed in
200  * Each block gets its own copy of pointers,
201  * and its unique 'value'
202  */
203  for (attr = 0; attr < nr_attrib; attr++) {
204  attrib = &attrib_p[attr];
205 
206  /* populate the unique per attrib
207  * with the code pointers and info
208  */
209  attrib->attr = attrib_spec[attr].attr;
210  attrib->show = attrib_spec[attr].show;
211  attrib->store = attrib_spec[attr].store;
212 
213  attrib->block = blk; /* up link */
214 
215  edac_dbg(4, "alloc-attrib=%p attrib_name='%s' attrib-spec=%p spec-name=%s\n",
216  attrib, attrib->attr.name,
217  &attrib_spec[attr],
218  attrib_spec[attr].attr.name
219  );
220  }
221  }
222  }
223 
224  /* Mark this instance as merely ALLOCATED */
225  dev_ctl->op_state = OP_ALLOC;
226 
227  /*
228  * Initialize the 'root' kobj for the edac_device controller
229  */
231  if (err) {
232  kfree(dev_ctl);
233  return NULL;
234  }
235 
236  /* at this point, the root kobj is valid, and in order to
237  * 'free' the object, then the function:
238  * edac_device_unregister_sysfs_main_kobj() must be called
239  * which will perform kobj unregistration and the actual free
240  * will occur during the kobject callback operation
241  */
242 
243  return dev_ctl;
244 }
246 
247 /*
248  * edac_device_free_ctl_info()
249  * frees the memory allocated by the edac_device_alloc_ctl_info()
250  * function
251  */
253 {
255 }
257 
258 /*
259  * find_edac_device_by_dev
260  * scans the edac_device list for a specific 'struct device *'
261  *
262  * lock to be held prior to call: device_ctls_mutex
263  *
264  * Return:
265  * pointer to control structure managing 'dev'
266  * NULL if not found on list
267  */
268 static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
269 {
270  struct edac_device_ctl_info *edac_dev;
271  struct list_head *item;
272 
273  edac_dbg(0, "\n");
274 
275  list_for_each(item, &edac_device_list) {
276  edac_dev = list_entry(item, struct edac_device_ctl_info, link);
277 
278  if (edac_dev->dev == dev)
279  return edac_dev;
280  }
281 
282  return NULL;
283 }
284 
285 /*
286  * add_edac_dev_to_global_list
287  * Before calling this function, caller must
288  * assign a unique value to edac_dev->dev_idx.
289  *
290  * lock to be held prior to call: device_ctls_mutex
291  *
292  * Return:
293  * 0 on success
294  * 1 on failure.
295  */
296 static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
297 {
298  struct list_head *item, *insert_before;
299  struct edac_device_ctl_info *rover;
300 
301  insert_before = &edac_device_list;
302 
303  /* Determine if already on the list */
304  rover = find_edac_device_by_dev(edac_dev->dev);
305  if (unlikely(rover != NULL))
306  goto fail0;
307 
308  /* Insert in ascending order by 'dev_idx', so find position */
309  list_for_each(item, &edac_device_list) {
310  rover = list_entry(item, struct edac_device_ctl_info, link);
311 
312  if (rover->dev_idx >= edac_dev->dev_idx) {
313  if (unlikely(rover->dev_idx == edac_dev->dev_idx))
314  goto fail1;
315 
316  insert_before = item;
317  break;
318  }
319  }
320 
321  list_add_tail_rcu(&edac_dev->link, insert_before);
322  return 0;
323 
324 fail0:
326  "%s (%s) %s %s already assigned %d\n",
327  dev_name(rover->dev), edac_dev_name(rover),
328  rover->mod_name, rover->ctl_name, rover->dev_idx);
329  return 1;
330 
331 fail1:
333  "bug in low-level driver: attempt to assign\n"
334  " duplicate dev_idx %d in %s()\n", rover->dev_idx,
335  __func__);
336  return 1;
337 }
338 
339 /*
340  * del_edac_device_from_global_list
341  */
342 static void del_edac_device_from_global_list(struct edac_device_ctl_info
343  *edac_device)
344 {
345  list_del_rcu(&edac_device->link);
346 
347  /* these are for safe removal of devices from global list while
348  * NMI handlers may be traversing list
349  */
350  synchronize_rcu();
351  INIT_LIST_HEAD(&edac_device->link);
352 }
353 
354 /*
355  * edac_device_workq_function
356  * performs the operation scheduled by a workq request
357  *
358  * this workq is embedded within an edac_device_ctl_info
359  * structure, that needs to be polled for possible error events.
360  *
361  * This operation is to acquire the list mutex lock
362  * (thus preventing insertation or deletion)
363  * and then call the device's poll function IFF this device is
364  * running polled and there is a poll function defined.
365  */
366 static void edac_device_workq_function(struct work_struct *work_req)
367 {
368  struct delayed_work *d_work = to_delayed_work(work_req);
369  struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);
370 
371  mutex_lock(&device_ctls_mutex);
372 
373  /* If we are being removed, bail out immediately */
374  if (edac_dev->op_state == OP_OFFLINE) {
375  mutex_unlock(&device_ctls_mutex);
376  return;
377  }
378 
379  /* Only poll controllers that are running polled and have a check */
380  if ((edac_dev->op_state == OP_RUNNING_POLL) &&
381  (edac_dev->edac_check != NULL)) {
382  edac_dev->edac_check(edac_dev);
383  }
384 
385  mutex_unlock(&device_ctls_mutex);
386 
387  /* Reschedule the workq for the next time period to start again
388  * if the number of msec is for 1 sec, then adjust to the next
389  * whole one second to save timers firing all over the period
390  * between integral seconds
391  */
392  if (edac_dev->poll_msec == 1000)
394  round_jiffies_relative(edac_dev->delay));
395  else
397  edac_dev->delay);
398 }
399 
400 /*
401  * edac_device_workq_setup
402  * initialize a workq item for this edac_device instance
403  * passing in the new delay period in msec
404  */
406  unsigned msec)
407 {
408  edac_dbg(0, "\n");
409 
410  /* take the arg 'msec' and set it into the control structure
411  * to used in the time period calculation
412  * then calc the number of jiffies that represents
413  */
414  edac_dev->poll_msec = msec;
415  edac_dev->delay = msecs_to_jiffies(msec);
416 
417  INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
418 
419  /* optimize here for the 1 second case, which will be normal value, to
420  * fire ON the 1 second time event. This helps reduce all sorts of
421  * timers firing on sub-second basis, while they are happy
422  * to fire together on the 1 second exactly
423  */
424  if (edac_dev->poll_msec == 1000)
426  round_jiffies_relative(edac_dev->delay));
427  else
429  edac_dev->delay);
430 }
431 
432 /*
433  * edac_device_workq_teardown
434  * stop the workq processing on this edac_dev
435  */
437 {
438  int status;
439 
440  status = cancel_delayed_work(&edac_dev->work);
441  if (status == 0) {
442  /* workq instance might be running, wait for it */
444  }
445 }
446 
447 /*
448  * edac_device_reset_delay_period
449  *
450  * need to stop any outstanding workq queued up at this time
451  * because we will be resetting the sleep time.
452  * Then restart the workq on the new delay
453  */
455  unsigned long value)
456 {
457  /* cancel the current workq request, without the mutex lock */
458  edac_device_workq_teardown(edac_dev);
459 
460  /* acquire the mutex before doing the workq setup */
461  mutex_lock(&device_ctls_mutex);
462 
463  /* restart the workq request, with new delay value */
464  edac_device_workq_setup(edac_dev, value);
465 
466  mutex_unlock(&device_ctls_mutex);
467 }
468 
469 /*
470  * edac_device_alloc_index: Allocate a unique device index number
471  *
472  * Return:
473  * allocated index number
474  */
476 {
477  static atomic_t device_indexes = ATOMIC_INIT(0);
478 
479  return atomic_inc_return(&device_indexes) - 1;
480 }
482 
495 {
496  edac_dbg(0, "\n");
497 
498 #ifdef CONFIG_EDAC_DEBUG
499  if (edac_debug_level >= 3)
500  edac_device_dump_device(edac_dev);
501 #endif
502  mutex_lock(&device_ctls_mutex);
503 
504  if (add_edac_dev_to_global_list(edac_dev))
505  goto fail0;
506 
507  /* set load time so that error rate can be tracked */
508  edac_dev->start_time = jiffies;
509 
510  /* create this instance's sysfs entries */
511  if (edac_device_create_sysfs(edac_dev)) {
513  "failed to create sysfs device\n");
514  goto fail1;
515  }
516 
517  /* If there IS a check routine, then we are running POLLED */
518  if (edac_dev->edac_check != NULL) {
519  /* This instance is NOW RUNNING */
520  edac_dev->op_state = OP_RUNNING_POLL;
521 
522  /*
523  * enable workq processing on this instance,
524  * default = 1000 msec
525  */
526  edac_device_workq_setup(edac_dev, 1000);
527  } else {
528  edac_dev->op_state = OP_RUNNING_INTERRUPT;
529  }
530 
531  /* Report action taken */
532  edac_device_printk(edac_dev, KERN_INFO,
533  "Giving out device to module '%s' controller "
534  "'%s': DEV '%s' (%s)\n",
535  edac_dev->mod_name,
536  edac_dev->ctl_name,
537  edac_dev_name(edac_dev),
538  edac_op_state_to_string(edac_dev->op_state));
539 
540  mutex_unlock(&device_ctls_mutex);
541  return 0;
542 
543 fail1:
544  /* Some error, so remove the entry from the lsit */
545  del_edac_device_from_global_list(edac_dev);
546 
547 fail0:
548  mutex_unlock(&device_ctls_mutex);
549  return 1;
550 }
552 
567 {
568  struct edac_device_ctl_info *edac_dev;
569 
570  edac_dbg(0, "\n");
571 
572  mutex_lock(&device_ctls_mutex);
573 
574  /* Find the structure on the list, if not there, then leave */
575  edac_dev = find_edac_device_by_dev(dev);
576  if (edac_dev == NULL) {
577  mutex_unlock(&device_ctls_mutex);
578  return NULL;
579  }
580 
581  /* mark this instance as OFFLINE */
582  edac_dev->op_state = OP_OFFLINE;
583 
584  /* deregister from global list */
585  del_edac_device_from_global_list(edac_dev);
586 
587  mutex_unlock(&device_ctls_mutex);
588 
589  /* clear workq processing on this instance */
590  edac_device_workq_teardown(edac_dev);
591 
592  /* Tear down the sysfs entries for this instance */
593  edac_device_remove_sysfs(edac_dev);
594 
596  "Removed device %d for %s %s: DEV %s\n",
597  edac_dev->dev_idx,
598  edac_dev->mod_name, edac_dev->ctl_name, edac_dev_name(edac_dev));
599 
600  return edac_dev;
601 }
603 
604 static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
605 {
606  return edac_dev->log_ce;
607 }
608 
609 static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
610 {
611  return edac_dev->log_ue;
612 }
613 
614 static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
615  *edac_dev)
616 {
617  return edac_dev->panic_on_ue;
618 }
619 
620 /*
621  * edac_device_handle_ce
622  * perform a common output and handling of an 'edac_dev' CE event
623  */
625  int inst_nr, int block_nr, const char *msg)
626 {
627  struct edac_device_instance *instance;
628  struct edac_device_block *block = NULL;
629 
630  if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
631  edac_device_printk(edac_dev, KERN_ERR,
632  "INTERNAL ERROR: 'instance' out of range "
633  "(%d >= %d)\n", inst_nr,
634  edac_dev->nr_instances);
635  return;
636  }
637 
638  instance = edac_dev->instances + inst_nr;
639 
640  if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
641  edac_device_printk(edac_dev, KERN_ERR,
642  "INTERNAL ERROR: instance %d 'block' "
643  "out of range (%d >= %d)\n",
644  inst_nr, block_nr,
645  instance->nr_blocks);
646  return;
647  }
648 
649  if (instance->nr_blocks > 0) {
650  block = instance->blocks + block_nr;
651  block->counters.ce_count++;
652  }
653 
654  /* Propagate the count up the 'totals' tree */
655  instance->counters.ce_count++;
656  edac_dev->counters.ce_count++;
657 
658  if (edac_device_get_log_ce(edac_dev))
660  "CE: %s instance: %s block: %s '%s'\n",
661  edac_dev->ctl_name, instance->name,
662  block ? block->name : "N/A", msg);
663 }
665 
666 /*
667  * edac_device_handle_ue
668  * perform a common output and handling of an 'edac_dev' UE event
669  */
671  int inst_nr, int block_nr, const char *msg)
672 {
673  struct edac_device_instance *instance;
674  struct edac_device_block *block = NULL;
675 
676  if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
677  edac_device_printk(edac_dev, KERN_ERR,
678  "INTERNAL ERROR: 'instance' out of range "
679  "(%d >= %d)\n", inst_nr,
680  edac_dev->nr_instances);
681  return;
682  }
683 
684  instance = edac_dev->instances + inst_nr;
685 
686  if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
687  edac_device_printk(edac_dev, KERN_ERR,
688  "INTERNAL ERROR: instance %d 'block' "
689  "out of range (%d >= %d)\n",
690  inst_nr, block_nr,
691  instance->nr_blocks);
692  return;
693  }
694 
695  if (instance->nr_blocks > 0) {
696  block = instance->blocks + block_nr;
697  block->counters.ue_count++;
698  }
699 
700  /* Propagate the count up the 'totals' tree */
701  instance->counters.ue_count++;
702  edac_dev->counters.ue_count++;
703 
704  if (edac_device_get_log_ue(edac_dev))
705  edac_device_printk(edac_dev, KERN_EMERG,
706  "UE: %s instance: %s block: %s '%s'\n",
707  edac_dev->ctl_name, instance->name,
708  block ? block->name : "N/A", msg);
709 
710  if (edac_device_get_panic_on_ue(edac_dev))
711  panic("EDAC %s: UE instance: %s block %s '%s'\n",
712  edac_dev->ctl_name, instance->name,
713  block ? block->name : "N/A", msg);
714 }