Loading...
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 <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 list
32 * is protected by the 'device_ctls_mutex' lock
33 */
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 1000
39
40#ifdef CONFIG_EDAC_DEBUG
41static 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 offset
55 */
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 struct edac_dev_sysfs_block_attribute *attrib_spec,
60 unsigned nr_attrib, int device_index)
61{
62 struct edac_dev_sysfs_block_attribute *dev_attrib, *attrib_p, *attrib;
63 struct edac_device_block *dev_blk, *blk_p, *blk;
64 struct edac_device_instance *dev_inst, *inst;
65 struct edac_device_ctl_info *dev_ctl;
66 unsigned instance, block, attr;
67 void *pvt;
68 int err;
69
70 edac_dbg(4, "instances=%d blocks=%d\n", nr_instances, nr_blocks);
71
72 dev_ctl = kzalloc(sizeof(struct edac_device_ctl_info), GFP_KERNEL);
73 if (!dev_ctl)
74 return NULL;
75
76 dev_inst = kcalloc(nr_instances, sizeof(struct edac_device_instance), GFP_KERNEL);
77 if (!dev_inst)
78 goto free;
79
80 dev_ctl->instances = dev_inst;
81
82 dev_blk = kcalloc(nr_instances * nr_blocks, sizeof(struct edac_device_block), GFP_KERNEL);
83 if (!dev_blk)
84 goto free;
85
86 dev_ctl->blocks = dev_blk;
87
88 if (nr_attrib) {
89 dev_attrib = kcalloc(nr_attrib, sizeof(struct edac_dev_sysfs_block_attribute),
90 GFP_KERNEL);
91 if (!dev_attrib)
92 goto free;
93
94 dev_ctl->attribs = dev_attrib;
95 }
96
97 if (pvt_sz) {
98 pvt = kzalloc(pvt_sz, GFP_KERNEL);
99 if (!pvt)
100 goto free;
101
102 dev_ctl->pvt_info = pvt;
103 }
104
105 dev_ctl->dev_idx = device_index;
106 dev_ctl->nr_instances = nr_instances;
107
108 /* Default logging of CEs and UEs */
109 dev_ctl->log_ce = 1;
110 dev_ctl->log_ue = 1;
111
112 /* Name of this edac device */
113 snprintf(dev_ctl->name, sizeof(dev_ctl->name),"%s", dev_name);
114
115 /* Initialize every Instance */
116 for (instance = 0; instance < nr_instances; instance++) {
117 inst = &dev_inst[instance];
118 inst->ctl = dev_ctl;
119 inst->nr_blocks = nr_blocks;
120 blk_p = &dev_blk[instance * nr_blocks];
121 inst->blocks = blk_p;
122
123 /* name of this instance */
124 snprintf(inst->name, sizeof(inst->name), "%s%u", dev_name, instance);
125
126 /* Initialize every block in each instance */
127 for (block = 0; block < nr_blocks; block++) {
128 blk = &blk_p[block];
129 blk->instance = inst;
130 snprintf(blk->name, sizeof(blk->name),
131 "%s%d", blk_name, block + off_val);
132
133 edac_dbg(4, "instance=%d inst_p=%p block=#%d block_p=%p name='%s'\n",
134 instance, inst, block, blk, blk->name);
135
136 /* if there are NO attributes OR no attribute pointer
137 * then continue on to next block iteration
138 */
139 if ((nr_attrib == 0) || (attrib_spec == NULL))
140 continue;
141
142 /* setup the attribute array for this block */
143 blk->nr_attribs = nr_attrib;
144 attrib_p = &dev_attrib[block*nr_instances*nr_attrib];
145 blk->block_attributes = attrib_p;
146
147 edac_dbg(4, "THIS BLOCK_ATTRIB=%p\n",
148 blk->block_attributes);
149
150 /* Initialize every user specified attribute in this
151 * block with the data the caller passed in
152 * Each block gets its own copy of pointers,
153 * and its unique 'value'
154 */
155 for (attr = 0; attr < nr_attrib; attr++) {
156 attrib = &attrib_p[attr];
157
158 /* populate the unique per attrib
159 * with the code pointers and info
160 */
161 attrib->attr = attrib_spec[attr].attr;
162 attrib->show = attrib_spec[attr].show;
163 attrib->store = attrib_spec[attr].store;
164
165 attrib->block = blk; /* up link */
166
167 edac_dbg(4, "alloc-attrib=%p attrib_name='%s' attrib-spec=%p spec-name=%s\n",
168 attrib, attrib->attr.name,
169 &attrib_spec[attr],
170 attrib_spec[attr].attr.name
171 );
172 }
173 }
174 }
175
176 /* Mark this instance as merely ALLOCATED */
177 dev_ctl->op_state = OP_ALLOC;
178
179 /*
180 * Initialize the 'root' kobj for the edac_device controller
181 */
182 err = edac_device_register_sysfs_main_kobj(dev_ctl);
183 if (err)
184 goto free;
185
186 /* at this point, the root kobj is valid, and in order to
187 * 'free' the object, then the function:
188 * edac_device_unregister_sysfs_main_kobj() must be called
189 * which will perform kobj unregistration and the actual free
190 * will occur during the kobject callback operation
191 */
192
193 return dev_ctl;
194
195free:
196 __edac_device_free_ctl_info(dev_ctl);
197
198 return NULL;
199}
200EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
201
202void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)
203{
204 edac_device_unregister_sysfs_main_kobj(ctl_info);
205}
206EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
207
208/*
209 * find_edac_device_by_dev
210 * scans the edac_device list for a specific 'struct device *'
211 *
212 * lock to be held prior to call: device_ctls_mutex
213 *
214 * Return:
215 * pointer to control structure managing 'dev'
216 * NULL if not found on list
217 */
218static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
219{
220 struct edac_device_ctl_info *edac_dev;
221 struct list_head *item;
222
223 edac_dbg(0, "\n");
224
225 list_for_each(item, &edac_device_list) {
226 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
227
228 if (edac_dev->dev == dev)
229 return edac_dev;
230 }
231
232 return NULL;
233}
234
235/*
236 * add_edac_dev_to_global_list
237 * Before calling this function, caller must
238 * assign a unique value to edac_dev->dev_idx.
239 *
240 * lock to be held prior to call: device_ctls_mutex
241 *
242 * Return:
243 * 0 on success
244 * 1 on failure.
245 */
246static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
247{
248 struct list_head *item, *insert_before;
249 struct edac_device_ctl_info *rover;
250
251 insert_before = &edac_device_list;
252
253 /* Determine if already on the list */
254 rover = find_edac_device_by_dev(edac_dev->dev);
255 if (unlikely(rover != NULL))
256 goto fail0;
257
258 /* Insert in ascending order by 'dev_idx', so find position */
259 list_for_each(item, &edac_device_list) {
260 rover = list_entry(item, struct edac_device_ctl_info, link);
261
262 if (rover->dev_idx >= edac_dev->dev_idx) {
263 if (unlikely(rover->dev_idx == edac_dev->dev_idx))
264 goto fail1;
265
266 insert_before = item;
267 break;
268 }
269 }
270
271 list_add_tail_rcu(&edac_dev->link, insert_before);
272 return 0;
273
274fail0:
275 edac_printk(KERN_WARNING, EDAC_MC,
276 "%s (%s) %s %s already assigned %d\n",
277 dev_name(rover->dev), edac_dev_name(rover),
278 rover->mod_name, rover->ctl_name, rover->dev_idx);
279 return 1;
280
281fail1:
282 edac_printk(KERN_WARNING, EDAC_MC,
283 "bug in low-level driver: attempt to assign\n"
284 " duplicate dev_idx %d in %s()\n", rover->dev_idx,
285 __func__);
286 return 1;
287}
288
289/*
290 * del_edac_device_from_global_list
291 */
292static void del_edac_device_from_global_list(struct edac_device_ctl_info
293 *edac_device)
294{
295 list_del_rcu(&edac_device->link);
296
297 /* these are for safe removal of devices from global list while
298 * NMI handlers may be traversing list
299 */
300 synchronize_rcu();
301 INIT_LIST_HEAD(&edac_device->link);
302}
303
304/*
305 * edac_device_workq_function
306 * performs the operation scheduled by a workq request
307 *
308 * this workq is embedded within an edac_device_ctl_info
309 * structure, that needs to be polled for possible error events.
310 *
311 * This operation is to acquire the list mutex lock
312 * (thus preventing insertation or deletion)
313 * and then call the device's poll function IFF this device is
314 * running polled and there is a poll function defined.
315 */
316static void edac_device_workq_function(struct work_struct *work_req)
317{
318 struct delayed_work *d_work = to_delayed_work(work_req);
319 struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);
320
321 mutex_lock(&device_ctls_mutex);
322
323 /* If we are being removed, bail out immediately */
324 if (edac_dev->op_state == OP_OFFLINE) {
325 mutex_unlock(&device_ctls_mutex);
326 return;
327 }
328
329 /* Only poll controllers that are running polled and have a check */
330 if ((edac_dev->op_state == OP_RUNNING_POLL) &&
331 (edac_dev->edac_check != NULL)) {
332 edac_dev->edac_check(edac_dev);
333 }
334
335 mutex_unlock(&device_ctls_mutex);
336
337 /* Reschedule the workq for the next time period to start again
338 * if the number of msec is for 1 sec, then adjust to the next
339 * whole one second to save timers firing all over the period
340 * between integral seconds
341 */
342 if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)
343 edac_queue_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));
344 else
345 edac_queue_work(&edac_dev->work, edac_dev->delay);
346}
347
348/*
349 * edac_device_workq_setup
350 * initialize a workq item for this edac_device instance
351 * passing in the new delay period in msec
352 */
353static void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
354 unsigned msec)
355{
356 edac_dbg(0, "\n");
357
358 /* take the arg 'msec' and set it into the control structure
359 * to used in the time period calculation
360 * then calc the number of jiffies that represents
361 */
362 edac_dev->poll_msec = msec;
363 edac_dev->delay = msecs_to_jiffies(msec);
364
365 INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
366
367 /* optimize here for the 1 second case, which will be normal value, to
368 * fire ON the 1 second time event. This helps reduce all sorts of
369 * timers firing on sub-second basis, while they are happy
370 * to fire together on the 1 second exactly
371 */
372 if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)
373 edac_queue_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));
374 else
375 edac_queue_work(&edac_dev->work, edac_dev->delay);
376}
377
378/*
379 * edac_device_workq_teardown
380 * stop the workq processing on this edac_dev
381 */
382static void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)
383{
384 if (!edac_dev->edac_check)
385 return;
386
387 edac_dev->op_state = OP_OFFLINE;
388
389 edac_stop_work(&edac_dev->work);
390}
391
392/*
393 * edac_device_reset_delay_period
394 *
395 * need to stop any outstanding workq queued up at this time
396 * because we will be resetting the sleep time.
397 * Then restart the workq on the new delay
398 */
399void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,
400 unsigned long msec)
401{
402 edac_dev->poll_msec = msec;
403 edac_dev->delay = msecs_to_jiffies(msec);
404
405 /* See comment in edac_device_workq_setup() above */
406 if (edac_dev->poll_msec == DEFAULT_POLL_INTERVAL)
407 edac_mod_work(&edac_dev->work, round_jiffies_relative(edac_dev->delay));
408 else
409 edac_mod_work(&edac_dev->work, edac_dev->delay);
410}
411
412int edac_device_alloc_index(void)
413{
414 static atomic_t device_indexes = ATOMIC_INIT(0);
415
416 return atomic_inc_return(&device_indexes) - 1;
417}
418EXPORT_SYMBOL_GPL(edac_device_alloc_index);
419
420int edac_device_add_device(struct edac_device_ctl_info *edac_dev)
421{
422 edac_dbg(0, "\n");
423
424#ifdef CONFIG_EDAC_DEBUG
425 if (edac_debug_level >= 3)
426 edac_device_dump_device(edac_dev);
427#endif
428 mutex_lock(&device_ctls_mutex);
429
430 if (add_edac_dev_to_global_list(edac_dev))
431 goto fail0;
432
433 /* set load time so that error rate can be tracked */
434 edac_dev->start_time = jiffies;
435
436 /* create this instance's sysfs entries */
437 if (edac_device_create_sysfs(edac_dev)) {
438 edac_device_printk(edac_dev, KERN_WARNING,
439 "failed to create sysfs device\n");
440 goto fail1;
441 }
442
443 /* If there IS a check routine, then we are running POLLED */
444 if (edac_dev->edac_check != NULL) {
445 /* This instance is NOW RUNNING */
446 edac_dev->op_state = OP_RUNNING_POLL;
447
448 edac_device_workq_setup(edac_dev, edac_dev->poll_msec ?: DEFAULT_POLL_INTERVAL);
449 } else {
450 edac_dev->op_state = OP_RUNNING_INTERRUPT;
451 }
452
453 /* Report action taken */
454 edac_device_printk(edac_dev, KERN_INFO,
455 "Giving out device to module %s controller %s: DEV %s (%s)\n",
456 edac_dev->mod_name, edac_dev->ctl_name, edac_dev->dev_name,
457 edac_op_state_to_string(edac_dev->op_state));
458
459 mutex_unlock(&device_ctls_mutex);
460 return 0;
461
462fail1:
463 /* Some error, so remove the entry from the lsit */
464 del_edac_device_from_global_list(edac_dev);
465
466fail0:
467 mutex_unlock(&device_ctls_mutex);
468 return 1;
469}
470EXPORT_SYMBOL_GPL(edac_device_add_device);
471
472struct edac_device_ctl_info *edac_device_del_device(struct device *dev)
473{
474 struct edac_device_ctl_info *edac_dev;
475
476 edac_dbg(0, "\n");
477
478 mutex_lock(&device_ctls_mutex);
479
480 /* Find the structure on the list, if not there, then leave */
481 edac_dev = find_edac_device_by_dev(dev);
482 if (edac_dev == NULL) {
483 mutex_unlock(&device_ctls_mutex);
484 return NULL;
485 }
486
487 /* mark this instance as OFFLINE */
488 edac_dev->op_state = OP_OFFLINE;
489
490 /* deregister from global list */
491 del_edac_device_from_global_list(edac_dev);
492
493 mutex_unlock(&device_ctls_mutex);
494
495 /* clear workq processing on this instance */
496 edac_device_workq_teardown(edac_dev);
497
498 /* Tear down the sysfs entries for this instance */
499 edac_device_remove_sysfs(edac_dev);
500
501 edac_printk(KERN_INFO, EDAC_MC,
502 "Removed device %d for %s %s: DEV %s\n",
503 edac_dev->dev_idx,
504 edac_dev->mod_name, edac_dev->ctl_name, edac_dev_name(edac_dev));
505
506 return edac_dev;
507}
508EXPORT_SYMBOL_GPL(edac_device_del_device);
509
510static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
511{
512 return edac_dev->log_ce;
513}
514
515static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
516{
517 return edac_dev->log_ue;
518}
519
520static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
521 *edac_dev)
522{
523 return edac_dev->panic_on_ue;
524}
525
526void edac_device_handle_ce_count(struct edac_device_ctl_info *edac_dev,
527 unsigned int count, int inst_nr, int block_nr,
528 const char *msg)
529{
530 struct edac_device_instance *instance;
531 struct edac_device_block *block = NULL;
532
533 if (!count)
534 return;
535
536 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
537 edac_device_printk(edac_dev, KERN_ERR,
538 "INTERNAL ERROR: 'instance' out of range "
539 "(%d >= %d)\n", inst_nr,
540 edac_dev->nr_instances);
541 return;
542 }
543
544 instance = edac_dev->instances + inst_nr;
545
546 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
547 edac_device_printk(edac_dev, KERN_ERR,
548 "INTERNAL ERROR: instance %d 'block' "
549 "out of range (%d >= %d)\n",
550 inst_nr, block_nr,
551 instance->nr_blocks);
552 return;
553 }
554
555 if (instance->nr_blocks > 0) {
556 block = instance->blocks + block_nr;
557 block->counters.ce_count += count;
558 }
559
560 /* Propagate the count up the 'totals' tree */
561 instance->counters.ce_count += count;
562 edac_dev->counters.ce_count += count;
563
564 if (edac_device_get_log_ce(edac_dev))
565 edac_device_printk(edac_dev, KERN_WARNING,
566 "CE: %s instance: %s block: %s count: %d '%s'\n",
567 edac_dev->ctl_name, instance->name,
568 block ? block->name : "N/A", count, msg);
569}
570EXPORT_SYMBOL_GPL(edac_device_handle_ce_count);
571
572void edac_device_handle_ue_count(struct edac_device_ctl_info *edac_dev,
573 unsigned int count, int inst_nr, int block_nr,
574 const char *msg)
575{
576 struct edac_device_instance *instance;
577 struct edac_device_block *block = NULL;
578
579 if (!count)
580 return;
581
582 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
583 edac_device_printk(edac_dev, KERN_ERR,
584 "INTERNAL ERROR: 'instance' out of range "
585 "(%d >= %d)\n", inst_nr,
586 edac_dev->nr_instances);
587 return;
588 }
589
590 instance = edac_dev->instances + inst_nr;
591
592 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
593 edac_device_printk(edac_dev, KERN_ERR,
594 "INTERNAL ERROR: instance %d 'block' "
595 "out of range (%d >= %d)\n",
596 inst_nr, block_nr,
597 instance->nr_blocks);
598 return;
599 }
600
601 if (instance->nr_blocks > 0) {
602 block = instance->blocks + block_nr;
603 block->counters.ue_count += count;
604 }
605
606 /* Propagate the count up the 'totals' tree */
607 instance->counters.ue_count += count;
608 edac_dev->counters.ue_count += count;
609
610 if (edac_device_get_log_ue(edac_dev))
611 edac_device_printk(edac_dev, KERN_EMERG,
612 "UE: %s instance: %s block: %s count: %d '%s'\n",
613 edac_dev->ctl_name, instance->name,
614 block ? block->name : "N/A", count, msg);
615
616 if (edac_device_get_panic_on_ue(edac_dev))
617 panic("EDAC %s: UE instance: %s block %s count: %d '%s'\n",
618 edac_dev->ctl_name, instance->name,
619 block ? block->name : "N/A", count, msg);
620}
621EXPORT_SYMBOL_GPL(edac_device_handle_ue_count);
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/jiffies.h>
24#include <linux/spinlock.h>
25#include <linux/list.h>
26#include <linux/sysdev.h>
27#include <linux/ctype.h>
28#include <linux/workqueue.h>
29#include <asm/uaccess.h>
30#include <asm/page.h>
31
32#include "edac_core.h"
33#include "edac_module.h"
34
35/* lock for the list: 'edac_device_list', manipulation of this list
36 * is protected by the 'device_ctls_mutex' lock
37 */
38static DEFINE_MUTEX(device_ctls_mutex);
39static LIST_HEAD(edac_device_list);
40
41#ifdef CONFIG_EDAC_DEBUG
42static void edac_device_dump_device(struct edac_device_ctl_info *edac_dev)
43{
44 debugf3("\tedac_dev = %p dev_idx=%d \n", edac_dev, edac_dev->dev_idx);
45 debugf4("\tedac_dev->edac_check = %p\n", edac_dev->edac_check);
46 debugf3("\tdev = %p\n", edac_dev->dev);
47 debugf3("\tmod_name:ctl_name = %s:%s\n",
48 edac_dev->mod_name, edac_dev->ctl_name);
49 debugf3("\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 struture
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 */
68struct edac_device_ctl_info *edac_device_alloc_ctl_info(
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;
84 int err;
85
86 debugf4("%s() instances=%d blocks=%d\n",
87 __func__, nr_instances, nr_blocks);
88
89 /* Calculate the size of memory we need to allocate AND
90 * determine the offsets of the various item arrays
91 * (instance,block,attrib) from the start of an allocated structure.
92 * We want the alignment of each item (instance,block,attrib)
93 * to be at least as stringent as what the compiler would
94 * provide if we could simply hardcode everything into a single struct.
95 */
96 dev_ctl = (struct edac_device_ctl_info *)NULL;
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(&dev_ctl[1], sizeof(*dev_inst));
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 dev_blk = edac_align_ptr(&dev_inst[nr_instances], sizeof(*dev_blk));
107
108 /* Calc the 'end' offset past the dev_blk array
109 * which will become the start of the attrib array, if any.
110 */
111 count = nr_instances * nr_blocks;
112 dev_attrib = edac_align_ptr(&dev_blk[count], sizeof(*dev_attrib));
113
114 /* Check for case of when an attribute array is specified */
115 if (nr_attrib > 0) {
116 /* calc how many nr_attrib we need */
117 count *= nr_attrib;
118
119 /* Calc the 'end' offset past the attributes array */
120 pvt = edac_align_ptr(&dev_attrib[count], sz_private);
121 } else {
122 /* no attribute array specificed */
123 pvt = edac_align_ptr(dev_attrib, sz_private);
124 }
125
126 /* 'pvt' now points to where the private data area is.
127 * At this point 'pvt' (like dev_inst,dev_blk and dev_attrib)
128 * is baselined at ZERO
129 */
130 total_size = ((unsigned long)pvt) + sz_private;
131
132 /* Allocate the amount of memory for the set of control structures */
133 dev_ctl = kzalloc(total_size, GFP_KERNEL);
134 if (dev_ctl == NULL)
135 return NULL;
136
137 /* Adjust pointers so they point within the actual memory we
138 * just allocated rather than an imaginary chunk of memory
139 * located at address 0.
140 * 'dev_ctl' points to REAL memory, while the others are
141 * ZERO based and thus need to be adjusted to point within
142 * the allocated memory.
143 */
144 dev_inst = (struct edac_device_instance *)
145 (((char *)dev_ctl) + ((unsigned long)dev_inst));
146 dev_blk = (struct edac_device_block *)
147 (((char *)dev_ctl) + ((unsigned long)dev_blk));
148 dev_attrib = (struct edac_dev_sysfs_block_attribute *)
149 (((char *)dev_ctl) + ((unsigned long)dev_attrib));
150 pvt = sz_private ? (((char *)dev_ctl) + ((unsigned long)pvt)) : NULL;
151
152 /* Begin storing the information into the control info structure */
153 dev_ctl->dev_idx = device_index;
154 dev_ctl->nr_instances = nr_instances;
155 dev_ctl->instances = dev_inst;
156 dev_ctl->pvt_info = pvt;
157
158 /* Default logging of CEs and UEs */
159 dev_ctl->log_ce = 1;
160 dev_ctl->log_ue = 1;
161
162 /* Name of this edac device */
163 snprintf(dev_ctl->name,sizeof(dev_ctl->name),"%s",edac_device_name);
164
165 debugf4("%s() edac_dev=%p next after end=%p\n",
166 __func__, dev_ctl, pvt + sz_private );
167
168 /* Initialize every Instance */
169 for (instance = 0; instance < nr_instances; instance++) {
170 inst = &dev_inst[instance];
171 inst->ctl = dev_ctl;
172 inst->nr_blocks = nr_blocks;
173 blk_p = &dev_blk[instance * nr_blocks];
174 inst->blocks = blk_p;
175
176 /* name of this instance */
177 snprintf(inst->name, sizeof(inst->name),
178 "%s%u", edac_device_name, instance);
179
180 /* Initialize every block in each instance */
181 for (block = 0; block < nr_blocks; block++) {
182 blk = &blk_p[block];
183 blk->instance = inst;
184 snprintf(blk->name, sizeof(blk->name),
185 "%s%d", edac_block_name, block+offset_value);
186
187 debugf4("%s() instance=%d inst_p=%p block=#%d "
188 "block_p=%p name='%s'\n",
189 __func__, instance, inst, block,
190 blk, blk->name);
191
192 /* if there are NO attributes OR no attribute pointer
193 * then continue on to next block iteration
194 */
195 if ((nr_attrib == 0) || (attrib_spec == NULL))
196 continue;
197
198 /* setup the attribute array for this block */
199 blk->nr_attribs = nr_attrib;
200 attrib_p = &dev_attrib[block*nr_instances*nr_attrib];
201 blk->block_attributes = attrib_p;
202
203 debugf4("%s() THIS BLOCK_ATTRIB=%p\n",
204 __func__, blk->block_attributes);
205
206 /* Initialize every user specified attribute in this
207 * block with the data the caller passed in
208 * Each block gets its own copy of pointers,
209 * and its unique 'value'
210 */
211 for (attr = 0; attr < nr_attrib; attr++) {
212 attrib = &attrib_p[attr];
213
214 /* populate the unique per attrib
215 * with the code pointers and info
216 */
217 attrib->attr = attrib_spec[attr].attr;
218 attrib->show = attrib_spec[attr].show;
219 attrib->store = attrib_spec[attr].store;
220
221 attrib->block = blk; /* up link */
222
223 debugf4("%s() alloc-attrib=%p attrib_name='%s' "
224 "attrib-spec=%p spec-name=%s\n",
225 __func__, attrib, attrib->attr.name,
226 &attrib_spec[attr],
227 attrib_spec[attr].attr.name
228 );
229 }
230 }
231 }
232
233 /* Mark this instance as merely ALLOCATED */
234 dev_ctl->op_state = OP_ALLOC;
235
236 /*
237 * Initialize the 'root' kobj for the edac_device controller
238 */
239 err = edac_device_register_sysfs_main_kobj(dev_ctl);
240 if (err) {
241 kfree(dev_ctl);
242 return NULL;
243 }
244
245 /* at this point, the root kobj is valid, and in order to
246 * 'free' the object, then the function:
247 * edac_device_unregister_sysfs_main_kobj() must be called
248 * which will perform kobj unregistration and the actual free
249 * will occur during the kobject callback operation
250 */
251
252 return dev_ctl;
253}
254EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
255
256/*
257 * edac_device_free_ctl_info()
258 * frees the memory allocated by the edac_device_alloc_ctl_info()
259 * function
260 */
261void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)
262{
263 edac_device_unregister_sysfs_main_kobj(ctl_info);
264}
265EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
266
267/*
268 * find_edac_device_by_dev
269 * scans the edac_device list for a specific 'struct device *'
270 *
271 * lock to be held prior to call: device_ctls_mutex
272 *
273 * Return:
274 * pointer to control structure managing 'dev'
275 * NULL if not found on list
276 */
277static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
278{
279 struct edac_device_ctl_info *edac_dev;
280 struct list_head *item;
281
282 debugf0("%s()\n", __func__);
283
284 list_for_each(item, &edac_device_list) {
285 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
286
287 if (edac_dev->dev == dev)
288 return edac_dev;
289 }
290
291 return NULL;
292}
293
294/*
295 * add_edac_dev_to_global_list
296 * Before calling this function, caller must
297 * assign a unique value to edac_dev->dev_idx.
298 *
299 * lock to be held prior to call: device_ctls_mutex
300 *
301 * Return:
302 * 0 on success
303 * 1 on failure.
304 */
305static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
306{
307 struct list_head *item, *insert_before;
308 struct edac_device_ctl_info *rover;
309
310 insert_before = &edac_device_list;
311
312 /* Determine if already on the list */
313 rover = find_edac_device_by_dev(edac_dev->dev);
314 if (unlikely(rover != NULL))
315 goto fail0;
316
317 /* Insert in ascending order by 'dev_idx', so find position */
318 list_for_each(item, &edac_device_list) {
319 rover = list_entry(item, struct edac_device_ctl_info, link);
320
321 if (rover->dev_idx >= edac_dev->dev_idx) {
322 if (unlikely(rover->dev_idx == edac_dev->dev_idx))
323 goto fail1;
324
325 insert_before = item;
326 break;
327 }
328 }
329
330 list_add_tail_rcu(&edac_dev->link, insert_before);
331 return 0;
332
333fail0:
334 edac_printk(KERN_WARNING, EDAC_MC,
335 "%s (%s) %s %s already assigned %d\n",
336 dev_name(rover->dev), edac_dev_name(rover),
337 rover->mod_name, rover->ctl_name, rover->dev_idx);
338 return 1;
339
340fail1:
341 edac_printk(KERN_WARNING, EDAC_MC,
342 "bug in low-level driver: attempt to assign\n"
343 " duplicate dev_idx %d in %s()\n", rover->dev_idx,
344 __func__);
345 return 1;
346}
347
348/*
349 * del_edac_device_from_global_list
350 */
351static void del_edac_device_from_global_list(struct edac_device_ctl_info
352 *edac_device)
353{
354 list_del_rcu(&edac_device->link);
355
356 /* these are for safe removal of devices from global list while
357 * NMI handlers may be traversing list
358 */
359 synchronize_rcu();
360 INIT_LIST_HEAD(&edac_device->link);
361}
362
363/*
364 * edac_device_workq_function
365 * performs the operation scheduled by a workq request
366 *
367 * this workq is embedded within an edac_device_ctl_info
368 * structure, that needs to be polled for possible error events.
369 *
370 * This operation is to acquire the list mutex lock
371 * (thus preventing insertation or deletion)
372 * and then call the device's poll function IFF this device is
373 * running polled and there is a poll function defined.
374 */
375static void edac_device_workq_function(struct work_struct *work_req)
376{
377 struct delayed_work *d_work = to_delayed_work(work_req);
378 struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);
379
380 mutex_lock(&device_ctls_mutex);
381
382 /* If we are being removed, bail out immediately */
383 if (edac_dev->op_state == OP_OFFLINE) {
384 mutex_unlock(&device_ctls_mutex);
385 return;
386 }
387
388 /* Only poll controllers that are running polled and have a check */
389 if ((edac_dev->op_state == OP_RUNNING_POLL) &&
390 (edac_dev->edac_check != NULL)) {
391 edac_dev->edac_check(edac_dev);
392 }
393
394 mutex_unlock(&device_ctls_mutex);
395
396 /* Reschedule the workq for the next time period to start again
397 * if the number of msec is for 1 sec, then adjust to the next
398 * whole one second to save timers fireing all over the period
399 * between integral seconds
400 */
401 if (edac_dev->poll_msec == 1000)
402 queue_delayed_work(edac_workqueue, &edac_dev->work,
403 round_jiffies_relative(edac_dev->delay));
404 else
405 queue_delayed_work(edac_workqueue, &edac_dev->work,
406 edac_dev->delay);
407}
408
409/*
410 * edac_device_workq_setup
411 * initialize a workq item for this edac_device instance
412 * passing in the new delay period in msec
413 */
414void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
415 unsigned msec)
416{
417 debugf0("%s()\n", __func__);
418
419 /* take the arg 'msec' and set it into the control structure
420 * to used in the time period calculation
421 * then calc the number of jiffies that represents
422 */
423 edac_dev->poll_msec = msec;
424 edac_dev->delay = msecs_to_jiffies(msec);
425
426 INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
427
428 /* optimize here for the 1 second case, which will be normal value, to
429 * fire ON the 1 second time event. This helps reduce all sorts of
430 * timers firing on sub-second basis, while they are happy
431 * to fire together on the 1 second exactly
432 */
433 if (edac_dev->poll_msec == 1000)
434 queue_delayed_work(edac_workqueue, &edac_dev->work,
435 round_jiffies_relative(edac_dev->delay));
436 else
437 queue_delayed_work(edac_workqueue, &edac_dev->work,
438 edac_dev->delay);
439}
440
441/*
442 * edac_device_workq_teardown
443 * stop the workq processing on this edac_dev
444 */
445void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)
446{
447 int status;
448
449 status = cancel_delayed_work(&edac_dev->work);
450 if (status == 0) {
451 /* workq instance might be running, wait for it */
452 flush_workqueue(edac_workqueue);
453 }
454}
455
456/*
457 * edac_device_reset_delay_period
458 *
459 * need to stop any outstanding workq queued up at this time
460 * because we will be resetting the sleep time.
461 * Then restart the workq on the new delay
462 */
463void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,
464 unsigned long value)
465{
466 /* cancel the current workq request, without the mutex lock */
467 edac_device_workq_teardown(edac_dev);
468
469 /* acquire the mutex before doing the workq setup */
470 mutex_lock(&device_ctls_mutex);
471
472 /* restart the workq request, with new delay value */
473 edac_device_workq_setup(edac_dev, value);
474
475 mutex_unlock(&device_ctls_mutex);
476}
477
478/*
479 * edac_device_alloc_index: Allocate a unique device index number
480 *
481 * Return:
482 * allocated index number
483 */
484int edac_device_alloc_index(void)
485{
486 static atomic_t device_indexes = ATOMIC_INIT(0);
487
488 return atomic_inc_return(&device_indexes) - 1;
489}
490EXPORT_SYMBOL_GPL(edac_device_alloc_index);
491
492/**
493 * edac_device_add_device: Insert the 'edac_dev' structure into the
494 * edac_device global list and create sysfs entries associated with
495 * edac_device structure.
496 * @edac_device: pointer to the edac_device structure to be added to the list
497 * 'edac_device' structure.
498 *
499 * Return:
500 * 0 Success
501 * !0 Failure
502 */
503int edac_device_add_device(struct edac_device_ctl_info *edac_dev)
504{
505 debugf0("%s()\n", __func__);
506
507#ifdef CONFIG_EDAC_DEBUG
508 if (edac_debug_level >= 3)
509 edac_device_dump_device(edac_dev);
510#endif
511 mutex_lock(&device_ctls_mutex);
512
513 if (add_edac_dev_to_global_list(edac_dev))
514 goto fail0;
515
516 /* set load time so that error rate can be tracked */
517 edac_dev->start_time = jiffies;
518
519 /* create this instance's sysfs entries */
520 if (edac_device_create_sysfs(edac_dev)) {
521 edac_device_printk(edac_dev, KERN_WARNING,
522 "failed to create sysfs device\n");
523 goto fail1;
524 }
525
526 /* If there IS a check routine, then we are running POLLED */
527 if (edac_dev->edac_check != NULL) {
528 /* This instance is NOW RUNNING */
529 edac_dev->op_state = OP_RUNNING_POLL;
530
531 /*
532 * enable workq processing on this instance,
533 * default = 1000 msec
534 */
535 edac_device_workq_setup(edac_dev, 1000);
536 } else {
537 edac_dev->op_state = OP_RUNNING_INTERRUPT;
538 }
539
540 /* Report action taken */
541 edac_device_printk(edac_dev, KERN_INFO,
542 "Giving out device to module '%s' controller "
543 "'%s': DEV '%s' (%s)\n",
544 edac_dev->mod_name,
545 edac_dev->ctl_name,
546 edac_dev_name(edac_dev),
547 edac_op_state_to_string(edac_dev->op_state));
548
549 mutex_unlock(&device_ctls_mutex);
550 return 0;
551
552fail1:
553 /* Some error, so remove the entry from the lsit */
554 del_edac_device_from_global_list(edac_dev);
555
556fail0:
557 mutex_unlock(&device_ctls_mutex);
558 return 1;
559}
560EXPORT_SYMBOL_GPL(edac_device_add_device);
561
562/**
563 * edac_device_del_device:
564 * Remove sysfs entries for specified edac_device structure and
565 * then remove edac_device structure from global list
566 *
567 * @pdev:
568 * Pointer to 'struct device' representing edac_device
569 * structure to remove.
570 *
571 * Return:
572 * Pointer to removed edac_device structure,
573 * OR NULL if device not found.
574 */
575struct edac_device_ctl_info *edac_device_del_device(struct device *dev)
576{
577 struct edac_device_ctl_info *edac_dev;
578
579 debugf0("%s()\n", __func__);
580
581 mutex_lock(&device_ctls_mutex);
582
583 /* Find the structure on the list, if not there, then leave */
584 edac_dev = find_edac_device_by_dev(dev);
585 if (edac_dev == NULL) {
586 mutex_unlock(&device_ctls_mutex);
587 return NULL;
588 }
589
590 /* mark this instance as OFFLINE */
591 edac_dev->op_state = OP_OFFLINE;
592
593 /* deregister from global list */
594 del_edac_device_from_global_list(edac_dev);
595
596 mutex_unlock(&device_ctls_mutex);
597
598 /* clear workq processing on this instance */
599 edac_device_workq_teardown(edac_dev);
600
601 /* Tear down the sysfs entries for this instance */
602 edac_device_remove_sysfs(edac_dev);
603
604 edac_printk(KERN_INFO, EDAC_MC,
605 "Removed device %d for %s %s: DEV %s\n",
606 edac_dev->dev_idx,
607 edac_dev->mod_name, edac_dev->ctl_name, edac_dev_name(edac_dev));
608
609 return edac_dev;
610}
611EXPORT_SYMBOL_GPL(edac_device_del_device);
612
613static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
614{
615 return edac_dev->log_ce;
616}
617
618static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
619{
620 return edac_dev->log_ue;
621}
622
623static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
624 *edac_dev)
625{
626 return edac_dev->panic_on_ue;
627}
628
629/*
630 * edac_device_handle_ce
631 * perform a common output and handling of an 'edac_dev' CE event
632 */
633void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
634 int inst_nr, int block_nr, const char *msg)
635{
636 struct edac_device_instance *instance;
637 struct edac_device_block *block = NULL;
638
639 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
640 edac_device_printk(edac_dev, KERN_ERR,
641 "INTERNAL ERROR: 'instance' out of range "
642 "(%d >= %d)\n", inst_nr,
643 edac_dev->nr_instances);
644 return;
645 }
646
647 instance = edac_dev->instances + inst_nr;
648
649 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
650 edac_device_printk(edac_dev, KERN_ERR,
651 "INTERNAL ERROR: instance %d 'block' "
652 "out of range (%d >= %d)\n",
653 inst_nr, block_nr,
654 instance->nr_blocks);
655 return;
656 }
657
658 if (instance->nr_blocks > 0) {
659 block = instance->blocks + block_nr;
660 block->counters.ce_count++;
661 }
662
663 /* Propagate the count up the 'totals' tree */
664 instance->counters.ce_count++;
665 edac_dev->counters.ce_count++;
666
667 if (edac_device_get_log_ce(edac_dev))
668 edac_device_printk(edac_dev, KERN_WARNING,
669 "CE: %s instance: %s block: %s '%s'\n",
670 edac_dev->ctl_name, instance->name,
671 block ? block->name : "N/A", msg);
672}
673EXPORT_SYMBOL_GPL(edac_device_handle_ce);
674
675/*
676 * edac_device_handle_ue
677 * perform a common output and handling of an 'edac_dev' UE event
678 */
679void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
680 int inst_nr, int block_nr, const char *msg)
681{
682 struct edac_device_instance *instance;
683 struct edac_device_block *block = NULL;
684
685 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
686 edac_device_printk(edac_dev, KERN_ERR,
687 "INTERNAL ERROR: 'instance' out of range "
688 "(%d >= %d)\n", inst_nr,
689 edac_dev->nr_instances);
690 return;
691 }
692
693 instance = edac_dev->instances + inst_nr;
694
695 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
696 edac_device_printk(edac_dev, KERN_ERR,
697 "INTERNAL ERROR: instance %d 'block' "
698 "out of range (%d >= %d)\n",
699 inst_nr, block_nr,
700 instance->nr_blocks);
701 return;
702 }
703
704 if (instance->nr_blocks > 0) {
705 block = instance->blocks + block_nr;
706 block->counters.ue_count++;
707 }
708
709 /* Propagate the count up the 'totals' tree */
710 instance->counters.ue_count++;
711 edac_dev->counters.ue_count++;
712
713 if (edac_device_get_log_ue(edac_dev))
714 edac_device_printk(edac_dev, KERN_EMERG,
715 "UE: %s instance: %s block: %s '%s'\n",
716 edac_dev->ctl_name, instance->name,
717 block ? block->name : "N/A", msg);
718
719 if (edac_device_get_panic_on_ue(edac_dev))
720 panic("EDAC %s: UE instance: %s block %s '%s'\n",
721 edac_dev->ctl_name, instance->name,
722 block ? block->name : "N/A", msg);
723}
724EXPORT_SYMBOL_GPL(edac_device_handle_ue);