aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/edac/edac_device.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/edac/edac_device.c')
-rw-r--r--drivers/edac/edac_device.c669
1 files changed, 669 insertions, 0 deletions
diff --git a/drivers/edac/edac_device.c b/drivers/edac/edac_device.c
new file mode 100644
index 000000000000..c579c498cc75
--- /dev/null
+++ b/drivers/edac/edac_device.c
@@ -0,0 +1,669 @@
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 */
35static DECLARE_MUTEX(device_ctls_mutex);
36static struct list_head edac_device_list = LIST_HEAD_INIT(edac_device_list);
37
38
39static inline void lock_device_list(void)
40{
41 down(&device_ctls_mutex);
42}
43
44static inline void unlock_device_list(void)
45{
46 up(&device_ctls_mutex);
47}
48
49
50#ifdef CONFIG_EDAC_DEBUG
51static void edac_device_dump_device(struct edac_device_ctl_info *edac_dev)
52{
53 debugf3("\tedac_dev = %p dev_idx=%d \n", edac_dev,edac_dev->dev_idx);
54 debugf4("\tedac_dev->edac_check = %p\n", edac_dev->edac_check);
55 debugf3("\tdev = %p\n", edac_dev->dev);
56 debugf3("\tmod_name:ctl_name = %s:%s\n",
57 edac_dev->mod_name, edac_dev->ctl_name);
58 debugf3("\tpvt_info = %p\n\n", edac_dev->pvt_info);
59}
60#endif /* CONFIG_EDAC_DEBUG */
61
62/*
63 * The alloc() and free() functions for the 'edac_device' control info
64 * structure. A MC driver will allocate one of these for each edac_device
65 * it is going to control/register with the EDAC CORE.
66 */
67struct edac_device_ctl_info *edac_device_alloc_ctl_info(
68 unsigned sz_private,
69 char *edac_device_name,
70 unsigned nr_instances,
71 char *edac_block_name,
72 unsigned nr_blocks,
73 unsigned offset_value,
74 struct edac_attrib_spec *attrib_spec,
75 unsigned nr_attribs)
76{
77 struct edac_device_ctl_info *dev_ctl;
78 struct edac_device_instance *dev_inst, *inst;
79 struct edac_device_block *dev_blk, *blk_p, *blk;
80 struct edac_attrib *dev_attrib, *attrib_p, *attrib;
81 unsigned total_size;
82 unsigned count;
83 unsigned instance, block, attr;
84 void *pvt;
85
86 debugf1("%s() instances=%d blocks=%d\n",
87 __func__,nr_instances,nr_blocks);
88
89 /* Figure out the offsets of the various items from the start of an
90 * ctl_info structure. We want the alignment of each item
91 * to be at least as stringent as what the compiler would
92 * provide if we could simply hardcode everything into a single struct.
93 */
94 dev_ctl = (struct edac_device_ctl_info *) 0;
95
96 /* Calc the 'end' offset past the ctl_info structure */
97 dev_inst = (struct edac_device_instance *)
98 edac_align_ptr(&dev_ctl[1],sizeof(*dev_inst));
99
100 /* Calc the 'end' offset past the instance array */
101 dev_blk = (struct edac_device_block *)
102 edac_align_ptr(&dev_inst[nr_instances],sizeof(*dev_blk));
103
104 /* Calc the 'end' offset past the dev_blk array */
105 count = nr_instances * nr_blocks;
106 dev_attrib = (struct edac_attrib *)
107 edac_align_ptr(&dev_blk[count],sizeof(*dev_attrib));
108
109 /* Check for case of NO attributes specified */
110 if (nr_attribs > 0)
111 count *= nr_attribs;
112
113 /* Calc the 'end' offset past the attributes array */
114 pvt = edac_align_ptr(&dev_attrib[count],sz_private);
115 total_size = ((unsigned long) pvt) + sz_private;
116
117 /* Allocate the amount of memory for the set of control structures */
118 if ((dev_ctl = kmalloc(total_size, GFP_KERNEL)) == NULL)
119 return NULL;
120
121 /* Adjust pointers so they point within the memory we just allocated
122 * rather than an imaginary chunk of memory located at address 0.
123 */
124 dev_inst = (struct edac_device_instance *)
125 (((char *) dev_ctl) + ((unsigned long) dev_inst));
126 dev_blk = (struct edac_device_block *)
127 (((char *) dev_ctl) + ((unsigned long) dev_blk));
128 dev_attrib = (struct edac_attrib *)
129 (((char *) dev_ctl) + ((unsigned long) dev_attrib));
130 pvt = sz_private ?
131 (((char *) dev_ctl) + ((unsigned long) pvt)) : NULL;
132
133 memset(dev_ctl, 0, total_size); /* clear all fields */
134 dev_ctl->nr_instances = nr_instances;
135 dev_ctl->instances = dev_inst;
136 dev_ctl->pvt_info = pvt;
137
138 /* Name of this edac device, ensure null terminated */
139 snprintf(dev_ctl->name,sizeof(dev_ctl->name),"%s", edac_device_name);
140 dev_ctl->name[sizeof(dev_ctl->name)-1] = '\0';
141
142 /* Initialize every Instance */
143 for (instance = 0; instance < nr_instances; instance++) {
144 inst = &dev_inst[instance];
145 inst->ctl = dev_ctl;
146 inst->nr_blocks = nr_blocks;
147 blk_p = &dev_blk[instance * nr_blocks];
148 inst->blocks = blk_p;
149
150 /* name of this instance */
151 snprintf(inst->name, sizeof(inst->name),
152 "%s%u", edac_device_name, instance);
153 inst->name[sizeof(inst->name)-1] = '\0';
154
155 /* Initialize every block in each instance */
156 for ( block = 0;
157 block < nr_blocks;
158 block++) {
159 blk = &blk_p[block];
160 blk->instance = inst;
161 blk->nr_attribs = nr_attribs;
162 attrib_p = &dev_attrib[block * nr_attribs];
163 blk->attribs = attrib_p;
164 snprintf(blk->name, sizeof(blk->name),
165 "%s%d", edac_block_name,block+1);
166 blk->name[sizeof(blk->name)-1] = '\0';
167
168 debugf1("%s() instance=%d block=%d name=%s\n",
169 __func__, instance,block,blk->name);
170
171 if (attrib_spec != NULL) {
172 /* when there is an attrib_spec passed int then
173 * Initialize every attrib of each block
174 */
175 for (attr = 0; attr < nr_attribs; attr++) {
176 attrib = &attrib_p[attr];
177 attrib->block = blk;
178
179 /* Link each attribute to the caller's
180 * spec entry, for name and type
181 */
182 attrib->spec = &attrib_spec[attr];
183 }
184 }
185 }
186 }
187
188 /* Mark this instance as merely ALLOCATED */
189 dev_ctl->op_state = OP_ALLOC;
190
191 return dev_ctl;
192}
193EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
194
195/*
196 * edac_device_free_ctl_info()
197 * frees the memory allocated by the edac_device_alloc_ctl_info()
198 * function
199 */
200void edac_device_free_ctl_info( struct edac_device_ctl_info *ctl_info) {
201 kfree(ctl_info);
202}
203EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
204
205
206
207/*
208 * find_edac_device_by_dev
209 * scans the edac_device list for a specific 'struct device *'
210 */
211static struct edac_device_ctl_info *
212find_edac_device_by_dev(struct device *dev)
213{
214 struct edac_device_ctl_info *edac_dev;
215 struct list_head *item;
216
217 debugf3("%s()\n", __func__);
218
219 list_for_each(item, &edac_device_list) {
220 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
221
222 if (edac_dev->dev == dev)
223 return edac_dev;
224 }
225
226 return NULL;
227}
228
229/*
230 * add_edac_dev_to_global_list
231 * Before calling this function, caller must
232 * assign a unique value to edac_dev->dev_idx.
233 * Return:
234 * 0 on success
235 * 1 on failure.
236 */
237static int add_edac_dev_to_global_list (struct edac_device_ctl_info *edac_dev)
238{
239 struct list_head *item, *insert_before;
240 struct edac_device_ctl_info *rover;
241
242 insert_before = &edac_device_list;
243
244 /* Determine if already on the list */
245 if (unlikely((rover = find_edac_device_by_dev(edac_dev->dev)) != NULL))
246 goto fail0;
247
248 /* Insert in ascending order by 'dev_idx', so find position */
249 list_for_each(item, &edac_device_list) {
250 rover = list_entry(item, struct edac_device_ctl_info, link);
251
252 if (rover->dev_idx >= edac_dev->dev_idx) {
253 if (unlikely(rover->dev_idx == edac_dev->dev_idx))
254 goto fail1;
255
256 insert_before = item;
257 break;
258 }
259 }
260
261 list_add_tail_rcu(&edac_dev->link, insert_before);
262 return 0;
263
264fail0:
265 edac_printk(KERN_WARNING, EDAC_MC,
266 "%s (%s) %s %s already assigned %d\n",
267 rover->dev->bus_id, dev_name(rover->dev),
268 rover->mod_name, rover->ctl_name, rover->dev_idx);
269 return 1;
270
271fail1:
272 edac_printk(KERN_WARNING, EDAC_MC,
273 "bug in low-level driver: attempt to assign\n"
274 " duplicate dev_idx %d in %s()\n", rover->dev_idx, __func__);
275 return 1;
276}
277
278/*
279 * complete_edac_device_list_del
280 */
281static void complete_edac_device_list_del(struct rcu_head *head)
282{
283 struct edac_device_ctl_info *edac_dev;
284
285 edac_dev = container_of(head, struct edac_device_ctl_info, rcu);
286 INIT_LIST_HEAD(&edac_dev->link);
287 complete(&edac_dev->complete);
288}
289
290/*
291 * del_edac_device_from_global_list
292 */
293static void del_edac_device_from_global_list(
294 struct edac_device_ctl_info *edac_device)
295{
296 list_del_rcu(&edac_device->link);
297 init_completion(&edac_device->complete);
298 call_rcu(&edac_device->rcu, complete_edac_device_list_del);
299 wait_for_completion(&edac_device->complete);
300}
301
302/**
303 * edac_device_find
304 * Search for a edac_device_ctl_info structure whose index is 'idx'.
305 *
306 * If found, return a pointer to the structure.
307 * Else return NULL.
308 *
309 * Caller must hold device_ctls_mutex.
310 */
311struct edac_device_ctl_info * edac_device_find(int idx)
312{
313 struct list_head *item;
314 struct edac_device_ctl_info *edac_dev;
315
316 /* Iterate over list, looking for exact match of ID */
317 list_for_each(item, &edac_device_list) {
318 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
319
320 if (edac_dev->dev_idx >= idx) {
321 if (edac_dev->dev_idx == idx)
322 return edac_dev;
323
324 /* not on list, so terminate early */
325 break;
326 }
327 }
328
329 return NULL;
330}
331EXPORT_SYMBOL(edac_device_find);
332
333
334/*
335 * edac_workq_function
336 * performs the operation scheduled by a workq request
337 */
338#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20))
339static void edac_workq_function(struct work_struct *work_req)
340{
341 struct delayed_work *d_work = (struct delayed_work*) work_req;
342 struct edac_device_ctl_info *edac_dev =
343 to_edac_device_ctl_work(d_work);
344#else
345static void edac_workq_function(void *ptr)
346{
347 struct edac_device_ctl_info *edac_dev =
348 (struct edac_device_ctl_info *) ptr;
349#endif
350
351 //debugf0("%s() here and running\n", __func__);
352 lock_device_list();
353
354 /* Only poll controllers that are running polled and have a check */
355 if ((edac_dev->op_state == OP_RUNNING_POLL) &&
356 (edac_dev->edac_check != NULL)) {
357 edac_dev->edac_check(edac_dev);
358 }
359
360 unlock_device_list();
361
362 /* Reschedule */
363 queue_delayed_work(edac_workqueue,&edac_dev->work, edac_dev->delay);
364}
365
366/*
367 * edac_workq_setup
368 * initialize a workq item for this edac_device instance
369 * passing in the new delay period in msec
370 */
371void edac_workq_setup(struct edac_device_ctl_info *edac_dev, unsigned msec)
372{
373 debugf0("%s()\n", __func__);
374
375 edac_dev->poll_msec = msec;
376 edac_device_calc_delay(edac_dev); /* Calc delay jiffies */
377
378#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20))
379 INIT_DELAYED_WORK(&edac_dev->work,edac_workq_function);
380#else
381 INIT_WORK(&edac_dev->work,edac_workq_function,edac_dev);
382#endif
383 queue_delayed_work(edac_workqueue,&edac_dev->work, edac_dev->delay);
384}
385
386/*
387 * edac_workq_teardown
388 * stop the workq processing on this edac_dev
389 */
390void edac_workq_teardown(struct edac_device_ctl_info *edac_dev)
391{
392 int status;
393
394 status = cancel_delayed_work(&edac_dev->work);
395 if (status == 0) {
396 /* workq instance might be running, wait for it */
397 flush_workqueue(edac_workqueue);
398 }
399}
400
401/*
402 * edac_device_reset_delay_period
403 */
404
405void edac_device_reset_delay_period(
406 struct edac_device_ctl_info *edac_dev,
407 unsigned long value)
408{
409 lock_device_list();
410
411 /* cancel the current workq request */
412 edac_workq_teardown(edac_dev);
413
414 /* restart the workq request, with new delay value */
415 edac_workq_setup(edac_dev, value);
416
417 unlock_device_list();
418}
419
420/*
421 * edac_op_state_toString(edac_dev)
422 */
423static char *edac_op_state_toString(struct edac_device_ctl_info *edac_dev)
424{
425 int opstate = edac_dev->op_state;
426
427 if (opstate == OP_RUNNING_POLL)
428 return "POLLED";
429 else if (opstate == OP_RUNNING_INTERRUPT)
430 return "INTERRUPT";
431 else if (opstate == OP_RUNNING_POLL_INTR)
432 return "POLL-INTR";
433 else if (opstate == OP_ALLOC)
434 return "ALLOC";
435 else if (opstate == OP_OFFLINE)
436 return "OFFLINE";
437
438 return "UNKNOWN";
439}
440
441/**
442 * edac_device_add_device: Insert the 'edac_dev' structure into the
443 * edac_device global list and create sysfs entries associated with
444 * edac_device structure.
445 * @edac_device: pointer to the edac_device structure to be added to the list
446 * @edac_idx: A unique numeric identifier to be assigned to the
447 * 'edac_device' structure.
448 *
449 * Return:
450 * 0 Success
451 * !0 Failure
452 */
453int edac_device_add_device(struct edac_device_ctl_info *edac_dev, int edac_idx)
454{
455 debugf0("%s()\n", __func__);
456
457 edac_dev->dev_idx = edac_idx;
458#ifdef CONFIG_EDAC_DEBUG
459 if (edac_debug_level >= 3)
460 edac_device_dump_device(edac_dev);
461#endif
462 lock_device_list();
463
464 if (add_edac_dev_to_global_list(edac_dev))
465 goto fail0;
466
467 /* set load time so that error rate can be tracked */
468 edac_dev->start_time = jiffies;
469
470 /* create this instance's sysfs entries */
471 if (edac_device_create_sysfs(edac_dev)) {
472 edac_device_printk(edac_dev, KERN_WARNING,
473 "failed to create sysfs device\n");
474 goto fail1;
475 }
476
477 /* If there IS a check routine, then we are running POLLED */
478 if (edac_dev->edac_check != NULL) {
479 /* This instance is NOW RUNNING */
480 edac_dev->op_state = OP_RUNNING_POLL;
481
482 /* enable workq processing on this instance, default = 1000 msec */
483 edac_workq_setup(edac_dev, 1000);
484 } else {
485 edac_dev->op_state = OP_RUNNING_INTERRUPT;
486 }
487
488
489 /* Report action taken */
490 edac_device_printk(edac_dev, KERN_INFO,
491 "Giving out device to module '%s' controller '%s': DEV '%s' (%s)\n",
492 edac_dev->mod_name,
493 edac_dev->ctl_name,
494 dev_name(edac_dev->dev),
495 edac_op_state_toString(edac_dev)
496 );
497
498 unlock_device_list();
499 return 0;
500
501fail1:
502 /* Some error, so remove the entry from the lsit */
503 del_edac_device_from_global_list(edac_dev);
504
505fail0:
506 unlock_device_list();
507 return 1;
508}
509EXPORT_SYMBOL_GPL(edac_device_add_device);
510
511/**
512 * edac_device_del_device:
513 * Remove sysfs entries for specified edac_device structure and
514 * then remove edac_device structure from global list
515 *
516 * @pdev:
517 * Pointer to 'struct device' representing edac_device
518 * structure to remove.
519 *
520 * Return:
521 * Pointer to removed edac_device structure,
522 * OR NULL if device not found.
523 */
524struct edac_device_ctl_info * edac_device_del_device(struct device *dev)
525{
526 struct edac_device_ctl_info *edac_dev;
527
528 debugf0("MC: %s()\n", __func__);
529
530 lock_device_list();
531
532 if ((edac_dev = find_edac_device_by_dev(dev)) == NULL) {
533 unlock_device_list();
534 return NULL;
535 }
536
537 /* mark this instance as OFFLINE */
538 edac_dev->op_state = OP_OFFLINE;
539
540 /* clear workq processing on this instance */
541 edac_workq_teardown(edac_dev);
542
543 /* Tear down the sysfs entries for this instance */
544 edac_device_remove_sysfs(edac_dev);
545
546 /* deregister from global list */
547 del_edac_device_from_global_list(edac_dev);
548
549 unlock_device_list();
550
551 edac_printk(KERN_INFO, EDAC_MC,
552 "Removed device %d for %s %s: DEV %s\n",
553 edac_dev->dev_idx,
554 edac_dev->mod_name,
555 edac_dev->ctl_name,
556 dev_name(edac_dev->dev));
557
558 return edac_dev;
559}
560EXPORT_SYMBOL_GPL(edac_device_del_device);
561
562
563static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
564{
565 return edac_dev->log_ce;
566}
567
568static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
569{
570 return edac_dev->log_ue;
571}
572
573static inline int edac_device_get_panic_on_ue(
574 struct edac_device_ctl_info *edac_dev)
575{
576 return edac_dev->panic_on_ue;
577}
578
579/*
580 * edac_device_handle_ce
581 * perform a common output and handling of an 'edac_dev' CE event
582 */
583void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
584 int inst_nr, int block_nr, const char *msg)
585{
586 struct edac_device_instance *instance;
587 struct edac_device_block *block = NULL;
588
589 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
590 edac_device_printk(edac_dev, KERN_ERR,
591 "INTERNAL ERROR: 'instance' out of range "
592 "(%d >= %d)\n", inst_nr, edac_dev->nr_instances);
593 return;
594 }
595
596 instance = edac_dev->instances + inst_nr;
597
598 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
599 edac_device_printk(edac_dev, KERN_ERR,
600 "INTERNAL ERROR: instance %d 'block' out of range "
601 "(%d >= %d)\n", inst_nr, block_nr, instance->nr_blocks);
602 return;
603 }
604
605 if (instance->nr_blocks > 0) {
606 block = instance->blocks + block_nr;
607 block->counters.ce_count++;
608 }
609
610 /* Propogate the count up the 'totals' tree */
611 instance->counters.ce_count++;
612 edac_dev->counters.ce_count++;
613
614 if (edac_device_get_log_ce(edac_dev))
615 edac_device_printk(edac_dev, KERN_WARNING,
616 "CE ctl: %s, instance: %s, block: %s: %s\n",
617 edac_dev->ctl_name, instance->name,
618 block ? block->name : "N/A", msg);
619}
620EXPORT_SYMBOL_GPL(edac_device_handle_ce);
621
622/*
623 * edac_device_handle_ue
624 * perform a common output and handling of an 'edac_dev' UE event
625 */
626void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
627 int inst_nr, int block_nr, const char *msg)
628{
629 struct edac_device_instance *instance;
630 struct edac_device_block *block = NULL;
631
632 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
633 edac_device_printk(edac_dev, KERN_ERR,
634 "INTERNAL ERROR: 'instance' out of range "
635 "(%d >= %d)\n", inst_nr, edac_dev->nr_instances);
636 return;
637 }
638
639 instance = edac_dev->instances + inst_nr;
640
641 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
642 edac_device_printk(edac_dev, KERN_ERR,
643 "INTERNAL ERROR: instance %d 'block' out of range "
644 "(%d >= %d)\n", inst_nr, block_nr, instance->nr_blocks);
645 return;
646 }
647
648 if (instance->nr_blocks > 0) {
649 block = instance->blocks + block_nr;
650 block->counters.ue_count++;
651 }
652
653 /* Propogate the count up the 'totals' tree */
654 instance->counters.ue_count++;
655 edac_dev->counters.ue_count++;
656
657 if (edac_device_get_log_ue(edac_dev))
658 edac_device_printk(edac_dev, KERN_EMERG,
659 "UE ctl: %s, instance: %s, block: %s: %s\n",
660 edac_dev->ctl_name, instance->name,
661 block ? block->name : "N/A", msg);
662
663 if (edac_device_get_panic_on_ue(edac_dev))
664 panic("EDAC %s: UE instance: %s, block %s: %s\n",
665 edac_dev->ctl_name, instance->name,
666 block ? block->name : "N/A", msg);
667}
668EXPORT_SYMBOL_GPL(edac_device_handle_ue);
669