4 * (C) 2007 www.douglaskthompson.com
6 * This file may be distributed under the terms of the
7 * GNU General Public License.
9 * Written by Doug Thompson <norsk5@xmission.com>
11 * edac_device API implementation
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>
31 #include "edac_core.h"
32 #include "edac_module.h"
34 /* lock for the list: 'edac_device_list', manipulation of this list
35 * is protected by the 'device_ctls_mutex' lock
37 static DEFINE_MUTEX(device_ctls_mutex);
38 static LIST_HEAD(edac_device_list);
40 #ifdef CONFIG_EDAC_DEBUG
41 static void edac_device_dump_device(struct edac_device_ctl_info *edac_dev)
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);
51 #endif /* CONFIG_EDAC_DEBUG */
55 * edac_device_alloc_ctl_info()
56 * Allocate a new edac device control info structure
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
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
66 * Within each instance is an array of 'nr_blocks' blockoffsets
68 struct edac_device_ctl_info *edac_device_alloc_ctl_info(
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,
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;
82 unsigned instance, block, attr;
86 edac_dbg(4, "instances=%d blocks=%d\n", nr_instances, nr_blocks);
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.
96 dev_ctl = edac_align_ptr(&p, sizeof(*dev_ctl), 1);
98 /* Calc the 'end' offset past end of ONE ctl_info structure
99 * which will become the start of the 'instance' array
101 dev_inst = edac_align_ptr(&p, sizeof(*dev_inst), nr_instances);
103 /* Calc the 'end' offset past the instance array within the ctl_info
104 * which will become the start of the block array
106 count = nr_instances * nr_blocks;
107 dev_blk = edac_align_ptr(&p, sizeof(*dev_blk), count);
109 /* Calc the 'end' offset past the dev_blk array
110 * which will become the start of the attrib array, if any.
112 /* calc how many nr_attrib we need */
115 dev_attrib = edac_align_ptr(&p, sizeof(*dev_attrib), count);
117 /* Calc the 'end' offset past the attributes array */
118 pvt = edac_align_ptr(&p, sz_private, 1);
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
124 total_size = ((unsigned long)pvt) + sz_private;
126 /* Allocate the amount of memory for the set of control structures */
127 dev_ctl = kzalloc(total_size, GFP_KERNEL);
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.
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;
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;
152 /* Default logging of CEs and UEs */
156 /* Name of this edac device */
157 snprintf(dev_ctl->name,sizeof(dev_ctl->name),"%s",edac_device_name);
159 edac_dbg(4, "edac_dev=%p next after end=%p\n",
160 dev_ctl, pvt + sz_private);
162 /* Initialize every Instance */
163 for (instance = 0; instance < nr_instances; instance++) {
164 inst = &dev_inst[instance];
166 inst->nr_blocks = nr_blocks;
167 blk_p = &dev_blk[instance * nr_blocks];
168 inst->blocks = blk_p;
170 /* name of this instance */
171 snprintf(inst->name, sizeof(inst->name),
172 "%s%u", edac_device_name, instance);
174 /* Initialize every block in each instance */
175 for (block = 0; block < nr_blocks; block++) {
177 blk->instance = inst;
178 snprintf(blk->name, sizeof(blk->name),
179 "%s%d", edac_block_name, block+offset_value);
181 edac_dbg(4, "instance=%d inst_p=%p block=#%d block_p=%p name='%s'\n",
182 instance, inst, block, blk, blk->name);
184 /* if there are NO attributes OR no attribute pointer
185 * then continue on to next block iteration
187 if ((nr_attrib == 0) || (attrib_spec == NULL))
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;
195 edac_dbg(4, "THIS BLOCK_ATTRIB=%p\n",
196 blk->block_attributes);
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'
203 for (attr = 0; attr < nr_attrib; attr++) {
204 attrib = &attrib_p[attr];
206 /* populate the unique per attrib
207 * with the code pointers and info
209 attrib->attr = attrib_spec[attr].attr;
210 attrib->show = attrib_spec[attr].show;
211 attrib->store = attrib_spec[attr].store;
213 attrib->block = blk; /* up link */
215 edac_dbg(4, "alloc-attrib=%p attrib_name='%s' attrib-spec=%p spec-name=%s\n",
216 attrib, attrib->attr.name,
218 attrib_spec[attr].attr.name
224 /* Mark this instance as merely ALLOCATED */
225 dev_ctl->op_state = OP_ALLOC;
228 * Initialize the 'root' kobj for the edac_device controller
230 err = edac_device_register_sysfs_main_kobj(dev_ctl);
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
245 EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
248 * edac_device_free_ctl_info()
249 * frees the memory allocated by the edac_device_alloc_ctl_info()
252 void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)
254 edac_device_unregister_sysfs_main_kobj(ctl_info);
256 EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
259 * find_edac_device_by_dev
260 * scans the edac_device list for a specific 'struct device *'
262 * lock to be held prior to call: device_ctls_mutex
265 * pointer to control structure managing 'dev'
266 * NULL if not found on list
268 static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
270 struct edac_device_ctl_info *edac_dev;
271 struct list_head *item;
275 list_for_each(item, &edac_device_list) {
276 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
278 if (edac_dev->dev == dev)
286 * add_edac_dev_to_global_list
287 * Before calling this function, caller must
288 * assign a unique value to edac_dev->dev_idx.
290 * lock to be held prior to call: device_ctls_mutex
296 static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
298 struct list_head *item, *insert_before;
299 struct edac_device_ctl_info *rover;
301 insert_before = &edac_device_list;
303 /* Determine if already on the list */
304 rover = find_edac_device_by_dev(edac_dev->dev);
305 if (unlikely(rover != NULL))
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);
312 if (rover->dev_idx >= edac_dev->dev_idx) {
313 if (unlikely(rover->dev_idx == edac_dev->dev_idx))
316 insert_before = item;
321 list_add_tail_rcu(&edac_dev->link, insert_before);
325 edac_printk(KERN_WARNING, EDAC_MC,
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);
332 edac_printk(KERN_WARNING, EDAC_MC,
333 "bug in low-level driver: attempt to assign\n"
334 " duplicate dev_idx %d in %s()\n", rover->dev_idx,
340 * del_edac_device_from_global_list
342 static void del_edac_device_from_global_list(struct edac_device_ctl_info
345 list_del_rcu(&edac_device->link);
347 /* these are for safe removal of devices from global list while
348 * NMI handlers may be traversing list
351 INIT_LIST_HEAD(&edac_device->link);
355 * edac_device_workq_function
356 * performs the operation scheduled by a workq request
358 * this workq is embedded within an edac_device_ctl_info
359 * structure, that needs to be polled for possible error events.
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.
366 static void edac_device_workq_function(struct work_struct *work_req)
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);
371 mutex_lock(&device_ctls_mutex);
373 /* If we are being removed, bail out immediately */
374 if (edac_dev->op_state == OP_OFFLINE) {
375 mutex_unlock(&device_ctls_mutex);
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);
385 mutex_unlock(&device_ctls_mutex);
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
392 if (edac_dev->poll_msec == 1000)
393 queue_delayed_work(edac_workqueue, &edac_dev->work,
394 round_jiffies_relative(edac_dev->delay));
396 queue_delayed_work(edac_workqueue, &edac_dev->work,
401 * edac_device_workq_setup
402 * initialize a workq item for this edac_device instance
403 * passing in the new delay period in msec
405 void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
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
414 edac_dev->poll_msec = msec;
415 edac_dev->delay = msecs_to_jiffies(msec);
417 INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
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
424 if (edac_dev->poll_msec == 1000)
425 queue_delayed_work(edac_workqueue, &edac_dev->work,
426 round_jiffies_relative(edac_dev->delay));
428 queue_delayed_work(edac_workqueue, &edac_dev->work,
433 * edac_device_workq_teardown
434 * stop the workq processing on this edac_dev
436 void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)
440 status = cancel_delayed_work(&edac_dev->work);
442 /* workq instance might be running, wait for it */
443 flush_workqueue(edac_workqueue);
448 * edac_device_reset_delay_period
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
454 void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,
457 /* cancel the current workq request, without the mutex lock */
458 edac_device_workq_teardown(edac_dev);
460 /* acquire the mutex before doing the workq setup */
461 mutex_lock(&device_ctls_mutex);
463 /* restart the workq request, with new delay value */
464 edac_device_workq_setup(edac_dev, value);
466 mutex_unlock(&device_ctls_mutex);
470 * edac_device_alloc_index: Allocate a unique device index number
473 * allocated index number
475 int edac_device_alloc_index(void)
477 static atomic_t device_indexes = ATOMIC_INIT(0);
479 return atomic_inc_return(&device_indexes) - 1;
481 EXPORT_SYMBOL_GPL(edac_device_alloc_index);
484 * edac_device_add_device: Insert the 'edac_dev' structure into the
485 * edac_device global list and create sysfs entries associated with
486 * edac_device structure.
487 * @edac_device: pointer to the edac_device structure to be added to the list
488 * 'edac_device' structure.
494 int edac_device_add_device(struct edac_device_ctl_info *edac_dev)
498 #ifdef CONFIG_EDAC_DEBUG
499 if (edac_debug_level >= 3)
500 edac_device_dump_device(edac_dev);
502 mutex_lock(&device_ctls_mutex);
504 if (add_edac_dev_to_global_list(edac_dev))
507 /* set load time so that error rate can be tracked */
508 edac_dev->start_time = jiffies;
510 /* create this instance's sysfs entries */
511 if (edac_device_create_sysfs(edac_dev)) {
512 edac_device_printk(edac_dev, KERN_WARNING,
513 "failed to create sysfs device\n");
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;
523 * enable workq processing on this instance,
524 * default = 1000 msec
526 edac_device_workq_setup(edac_dev, 1000);
528 edac_dev->op_state = OP_RUNNING_INTERRUPT;
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",
537 edac_dev_name(edac_dev),
538 edac_op_state_to_string(edac_dev->op_state));
540 mutex_unlock(&device_ctls_mutex);
544 /* Some error, so remove the entry from the lsit */
545 del_edac_device_from_global_list(edac_dev);
548 mutex_unlock(&device_ctls_mutex);
551 EXPORT_SYMBOL_GPL(edac_device_add_device);
554 * edac_device_del_device:
555 * Remove sysfs entries for specified edac_device structure and
556 * then remove edac_device structure from global list
559 * Pointer to 'struct device' representing edac_device
560 * structure to remove.
563 * Pointer to removed edac_device structure,
564 * OR NULL if device not found.
566 struct edac_device_ctl_info *edac_device_del_device(struct device *dev)
568 struct edac_device_ctl_info *edac_dev;
572 mutex_lock(&device_ctls_mutex);
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);
581 /* mark this instance as OFFLINE */
582 edac_dev->op_state = OP_OFFLINE;
584 /* deregister from global list */
585 del_edac_device_from_global_list(edac_dev);
587 mutex_unlock(&device_ctls_mutex);
589 /* clear workq processing on this instance */
590 edac_device_workq_teardown(edac_dev);
592 /* Tear down the sysfs entries for this instance */
593 edac_device_remove_sysfs(edac_dev);
595 edac_printk(KERN_INFO, EDAC_MC,
596 "Removed device %d for %s %s: DEV %s\n",
598 edac_dev->mod_name, edac_dev->ctl_name, edac_dev_name(edac_dev));
602 EXPORT_SYMBOL_GPL(edac_device_del_device);
604 static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
606 return edac_dev->log_ce;
609 static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
611 return edac_dev->log_ue;
614 static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
617 return edac_dev->panic_on_ue;
621 * edac_device_handle_ce
622 * perform a common output and handling of an 'edac_dev' CE event
624 void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
625 int inst_nr, int block_nr, const char *msg)
627 struct edac_device_instance *instance;
628 struct edac_device_block *block = NULL;
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);
638 instance = edac_dev->instances + inst_nr;
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",
645 instance->nr_blocks);
649 if (instance->nr_blocks > 0) {
650 block = instance->blocks + block_nr;
651 block->counters.ce_count++;
654 /* Propagate the count up the 'totals' tree */
655 instance->counters.ce_count++;
656 edac_dev->counters.ce_count++;
658 if (edac_device_get_log_ce(edac_dev))
659 edac_device_printk(edac_dev, KERN_WARNING,
660 "CE: %s instance: %s block: %s '%s'\n",
661 edac_dev->ctl_name, instance->name,
662 block ? block->name : "N/A", msg);
664 EXPORT_SYMBOL_GPL(edac_device_handle_ce);
667 * edac_device_handle_ue
668 * perform a common output and handling of an 'edac_dev' UE event
670 void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
671 int inst_nr, int block_nr, const char *msg)
673 struct edac_device_instance *instance;
674 struct edac_device_block *block = NULL;
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);
684 instance = edac_dev->instances + inst_nr;
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",
691 instance->nr_blocks);
695 if (instance->nr_blocks > 0) {
696 block = instance->blocks + block_nr;
697 block->counters.ue_count++;
700 /* Propagate the count up the 'totals' tree */
701 instance->counters.ue_count++;
702 edac_dev->counters.ue_count++;
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);
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);
715 EXPORT_SYMBOL_GPL(edac_device_handle_ue);