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/sysdev.h>
27 #include <linux/ctype.h>
28 #include <linux/workqueue.h>
29 #include <asm/uaccess.h>
32 #include "edac_core.h"
33 #include "edac_module.h"
35 /* lock for the list: 'edac_device_list', manipulation of this list
36 * is protected by the 'device_ctls_mutex' lock
38 static DEFINE_MUTEX(device_ctls_mutex);
39 static struct list_head edac_device_list = LIST_HEAD_INIT(edac_device_list);
41 #ifdef CONFIG_EDAC_DEBUG
42 static void edac_device_dump_device(struct edac_device_ctl_info *edac_dev)
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);
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 struture
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 debugf4("%s() instances=%d blocks=%d\n",
87 __func__, nr_instances, nr_blocks);
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.
96 dev_ctl = (struct edac_device_ctl_info *)NULL;
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(&dev_ctl[1], sizeof(*dev_inst));
103 /* Calc the 'end' offset past the instance array within the ctl_info
104 * which will become the start of the block array
106 dev_blk = edac_align_ptr(&dev_inst[nr_instances], sizeof(*dev_blk));
108 /* Calc the 'end' offset past the dev_blk array
109 * which will become the start of the attrib array, if any.
111 count = nr_instances * nr_blocks;
112 dev_attrib = edac_align_ptr(&dev_blk[count], sizeof(*dev_attrib));
114 /* Check for case of when an attribute array is specified */
116 /* calc how many nr_attrib we need */
119 /* Calc the 'end' offset past the attributes array */
120 pvt = edac_align_ptr(&dev_attrib[count], sz_private);
122 /* no attribute array specificed */
123 pvt = edac_align_ptr(dev_attrib, sz_private);
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
130 total_size = ((unsigned long)pvt) + sz_private;
132 /* Allocate the amount of memory for the set of control structures */
133 dev_ctl = kzalloc(total_size, GFP_KERNEL);
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.
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;
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;
158 /* Name of this edac device */
159 snprintf(dev_ctl->name,sizeof(dev_ctl->name),"%s",edac_device_name);
161 debugf4("%s() edac_dev=%p next after end=%p\n",
162 __func__, dev_ctl, pvt + sz_private );
164 /* Initialize every Instance */
165 for (instance = 0; instance < nr_instances; instance++) {
166 inst = &dev_inst[instance];
168 inst->nr_blocks = nr_blocks;
169 blk_p = &dev_blk[instance * nr_blocks];
170 inst->blocks = blk_p;
172 /* name of this instance */
173 snprintf(inst->name, sizeof(inst->name),
174 "%s%u", edac_device_name, instance);
176 /* Initialize every block in each instance */
177 for (block = 0; block < nr_blocks; block++) {
179 blk->instance = inst;
180 snprintf(blk->name, sizeof(blk->name),
181 "%s%d", edac_block_name, block+offset_value);
183 debugf4("%s() instance=%d inst_p=%p block=#%d "
184 "block_p=%p name='%s'\n",
185 __func__, instance, inst, block,
188 /* if there are NO attributes OR no attribute pointer
189 * then continue on to next block iteration
191 if ((nr_attrib == 0) || (attrib_spec == NULL))
194 /* setup the attribute array for this block */
195 blk->nr_attribs = nr_attrib;
196 attrib_p = &dev_attrib[block*nr_instances*nr_attrib];
197 blk->block_attributes = attrib_p;
199 debugf4("%s() THIS BLOCK_ATTRIB=%p\n",
200 __func__, blk->block_attributes);
202 /* Initialize every user specified attribute in this
203 * block with the data the caller passed in
204 * Each block gets its own copy of pointers,
205 * and its unique 'value'
207 for (attr = 0; attr < nr_attrib; attr++) {
208 attrib = &attrib_p[attr];
210 /* populate the unique per attrib
211 * with the code pointers and info
213 attrib->attr = attrib_spec[attr].attr;
214 attrib->show = attrib_spec[attr].show;
215 attrib->store = attrib_spec[attr].store;
217 attrib->block = blk; /* up link */
219 debugf4("%s() alloc-attrib=%p attrib_name='%s' "
220 "attrib-spec=%p spec-name=%s\n",
221 __func__, attrib, attrib->attr.name,
223 attrib_spec[attr].attr.name
229 /* Mark this instance as merely ALLOCATED */
230 dev_ctl->op_state = OP_ALLOC;
233 * Initialize the 'root' kobj for the edac_device controller
235 err = edac_device_register_sysfs_main_kobj(dev_ctl);
241 /* at this point, the root kobj is valid, and in order to
242 * 'free' the object, then the function:
243 * edac_device_unregister_sysfs_main_kobj() must be called
244 * which will perform kobj unregistration and the actual free
245 * will occur during the kobject callback operation
250 EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
253 * edac_device_free_ctl_info()
254 * frees the memory allocated by the edac_device_alloc_ctl_info()
257 void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)
259 edac_device_unregister_sysfs_main_kobj(ctl_info);
261 EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
264 * find_edac_device_by_dev
265 * scans the edac_device list for a specific 'struct device *'
267 * lock to be held prior to call: device_ctls_mutex
270 * pointer to control structure managing 'dev'
271 * NULL if not found on list
273 static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
275 struct edac_device_ctl_info *edac_dev;
276 struct list_head *item;
278 debugf0("%s()\n", __func__);
280 list_for_each(item, &edac_device_list) {
281 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
283 if (edac_dev->dev == dev)
291 * add_edac_dev_to_global_list
292 * Before calling this function, caller must
293 * assign a unique value to edac_dev->dev_idx.
295 * lock to be held prior to call: device_ctls_mutex
301 static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
303 struct list_head *item, *insert_before;
304 struct edac_device_ctl_info *rover;
306 insert_before = &edac_device_list;
308 /* Determine if already on the list */
309 rover = find_edac_device_by_dev(edac_dev->dev);
310 if (unlikely(rover != NULL))
313 /* Insert in ascending order by 'dev_idx', so find position */
314 list_for_each(item, &edac_device_list) {
315 rover = list_entry(item, struct edac_device_ctl_info, link);
317 if (rover->dev_idx >= edac_dev->dev_idx) {
318 if (unlikely(rover->dev_idx == edac_dev->dev_idx))
321 insert_before = item;
326 list_add_tail_rcu(&edac_dev->link, insert_before);
330 edac_printk(KERN_WARNING, EDAC_MC,
331 "%s (%s) %s %s already assigned %d\n",
332 rover->dev->bus_id, dev_name(rover),
333 rover->mod_name, rover->ctl_name, rover->dev_idx);
337 edac_printk(KERN_WARNING, EDAC_MC,
338 "bug in low-level driver: attempt to assign\n"
339 " duplicate dev_idx %d in %s()\n", rover->dev_idx,
345 * complete_edac_device_list_del
347 * callback function when reference count is zero
349 static void complete_edac_device_list_del(struct rcu_head *head)
351 struct edac_device_ctl_info *edac_dev;
353 edac_dev = container_of(head, struct edac_device_ctl_info, rcu);
354 INIT_LIST_HEAD(&edac_dev->link);
355 complete(&edac_dev->removal_complete);
359 * del_edac_device_from_global_list
361 * remove the RCU, setup for a callback call,
362 * then wait for the callback to occur
364 static void del_edac_device_from_global_list(struct edac_device_ctl_info
367 list_del_rcu(&edac_device->link);
369 init_completion(&edac_device->removal_complete);
370 call_rcu(&edac_device->rcu, complete_edac_device_list_del);
371 wait_for_completion(&edac_device->removal_complete);
376 * Search for a edac_device_ctl_info structure whose index is 'idx'.
378 * If found, return a pointer to the structure.
381 * Caller must hold device_ctls_mutex.
383 struct edac_device_ctl_info *edac_device_find(int idx)
385 struct list_head *item;
386 struct edac_device_ctl_info *edac_dev;
388 /* Iterate over list, looking for exact match of ID */
389 list_for_each(item, &edac_device_list) {
390 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
392 if (edac_dev->dev_idx >= idx) {
393 if (edac_dev->dev_idx == idx)
396 /* not on list, so terminate early */
403 EXPORT_SYMBOL_GPL(edac_device_find);
406 * edac_device_workq_function
407 * performs the operation scheduled by a workq request
409 * this workq is embedded within an edac_device_ctl_info
410 * structure, that needs to be polled for possible error events.
412 * This operation is to acquire the list mutex lock
413 * (thus preventing insertation or deletion)
414 * and then call the device's poll function IFF this device is
415 * running polled and there is a poll function defined.
417 static void edac_device_workq_function(struct work_struct *work_req)
419 struct delayed_work *d_work = (struct delayed_work *)work_req;
420 struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);
422 mutex_lock(&device_ctls_mutex);
424 /* Only poll controllers that are running polled and have a check */
425 if ((edac_dev->op_state == OP_RUNNING_POLL) &&
426 (edac_dev->edac_check != NULL)) {
427 edac_dev->edac_check(edac_dev);
430 mutex_unlock(&device_ctls_mutex);
432 /* Reschedule the workq for the next time period to start again
433 * if the number of msec is for 1 sec, then adjust to the next
434 * whole one second to save timers fireing all over the period
435 * between integral seconds
437 if (edac_dev->poll_msec == 1000)
438 queue_delayed_work(edac_workqueue, &edac_dev->work,
439 round_jiffies(edac_dev->delay));
441 queue_delayed_work(edac_workqueue, &edac_dev->work,
446 * edac_device_workq_setup
447 * initialize a workq item for this edac_device instance
448 * passing in the new delay period in msec
450 void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
453 debugf0("%s()\n", __func__);
455 /* take the arg 'msec' and set it into the control structure
456 * to used in the time period calculation
457 * then calc the number of jiffies that represents
459 edac_dev->poll_msec = msec;
460 edac_dev->delay = msecs_to_jiffies(msec);
462 INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
464 /* optimize here for the 1 second case, which will be normal value, to
465 * fire ON the 1 second time event. This helps reduce all sorts of
466 * timers firing on sub-second basis, while they are happy
467 * to fire together on the 1 second exactly
469 if (edac_dev->poll_msec == 1000)
470 queue_delayed_work(edac_workqueue, &edac_dev->work,
471 round_jiffies(edac_dev->delay));
473 queue_delayed_work(edac_workqueue, &edac_dev->work,
478 * edac_device_workq_teardown
479 * stop the workq processing on this edac_dev
481 void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)
485 status = cancel_delayed_work(&edac_dev->work);
487 /* workq instance might be running, wait for it */
488 flush_workqueue(edac_workqueue);
493 * edac_device_reset_delay_period
495 * need to stop any outstanding workq queued up at this time
496 * because we will be resetting the sleep time.
497 * Then restart the workq on the new delay
499 void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,
502 /* cancel the current workq request, without the mutex lock */
503 edac_device_workq_teardown(edac_dev);
505 /* acquire the mutex before doing the workq setup */
506 mutex_lock(&device_ctls_mutex);
508 /* restart the workq request, with new delay value */
509 edac_device_workq_setup(edac_dev, value);
511 mutex_unlock(&device_ctls_mutex);
515 * edac_device_add_device: Insert the 'edac_dev' structure into the
516 * edac_device global list and create sysfs entries associated with
517 * edac_device structure.
518 * @edac_device: pointer to the edac_device structure to be added to the list
519 * 'edac_device' structure.
525 int edac_device_add_device(struct edac_device_ctl_info *edac_dev)
527 debugf0("%s()\n", __func__);
529 #ifdef CONFIG_EDAC_DEBUG
530 if (edac_debug_level >= 3)
531 edac_device_dump_device(edac_dev);
533 mutex_lock(&device_ctls_mutex);
535 if (add_edac_dev_to_global_list(edac_dev))
538 /* set load time so that error rate can be tracked */
539 edac_dev->start_time = jiffies;
541 /* create this instance's sysfs entries */
542 if (edac_device_create_sysfs(edac_dev)) {
543 edac_device_printk(edac_dev, KERN_WARNING,
544 "failed to create sysfs device\n");
548 /* If there IS a check routine, then we are running POLLED */
549 if (edac_dev->edac_check != NULL) {
550 /* This instance is NOW RUNNING */
551 edac_dev->op_state = OP_RUNNING_POLL;
554 * enable workq processing on this instance,
555 * default = 1000 msec
557 edac_device_workq_setup(edac_dev, 1000);
559 edac_dev->op_state = OP_RUNNING_INTERRUPT;
562 /* Report action taken */
563 edac_device_printk(edac_dev, KERN_INFO,
564 "Giving out device to module '%s' controller "
565 "'%s': DEV '%s' (%s)\n",
569 edac_op_state_to_string(edac_dev->op_state));
571 mutex_unlock(&device_ctls_mutex);
575 /* Some error, so remove the entry from the lsit */
576 del_edac_device_from_global_list(edac_dev);
579 mutex_unlock(&device_ctls_mutex);
582 EXPORT_SYMBOL_GPL(edac_device_add_device);
585 * edac_device_del_device:
586 * Remove sysfs entries for specified edac_device structure and
587 * then remove edac_device structure from global list
590 * Pointer to 'struct device' representing edac_device
591 * structure to remove.
594 * Pointer to removed edac_device structure,
595 * OR NULL if device not found.
597 struct edac_device_ctl_info *edac_device_del_device(struct device *dev)
599 struct edac_device_ctl_info *edac_dev;
601 debugf0("%s()\n", __func__);
603 mutex_lock(&device_ctls_mutex);
605 /* Find the structure on the list, if not there, then leave */
606 edac_dev = find_edac_device_by_dev(dev);
607 if (edac_dev == NULL) {
608 mutex_unlock(&device_ctls_mutex);
612 /* mark this instance as OFFLINE */
613 edac_dev->op_state = OP_OFFLINE;
615 /* clear workq processing on this instance */
616 edac_device_workq_teardown(edac_dev);
618 /* deregister from global list */
619 del_edac_device_from_global_list(edac_dev);
621 mutex_unlock(&device_ctls_mutex);
623 /* Tear down the sysfs entries for this instance */
624 edac_device_remove_sysfs(edac_dev);
626 edac_printk(KERN_INFO, EDAC_MC,
627 "Removed device %d for %s %s: DEV %s\n",
629 edac_dev->mod_name, edac_dev->ctl_name, dev_name(edac_dev));
633 EXPORT_SYMBOL_GPL(edac_device_del_device);
635 static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
637 return edac_dev->log_ce;
640 static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
642 return edac_dev->log_ue;
645 static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
648 return edac_dev->panic_on_ue;
652 * edac_device_handle_ce
653 * perform a common output and handling of an 'edac_dev' CE event
655 void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
656 int inst_nr, int block_nr, const char *msg)
658 struct edac_device_instance *instance;
659 struct edac_device_block *block = NULL;
661 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
662 edac_device_printk(edac_dev, KERN_ERR,
663 "INTERNAL ERROR: 'instance' out of range "
664 "(%d >= %d)\n", inst_nr,
665 edac_dev->nr_instances);
669 instance = edac_dev->instances + inst_nr;
671 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
672 edac_device_printk(edac_dev, KERN_ERR,
673 "INTERNAL ERROR: instance %d 'block' "
674 "out of range (%d >= %d)\n",
676 instance->nr_blocks);
680 if (instance->nr_blocks > 0) {
681 block = instance->blocks + block_nr;
682 block->counters.ce_count++;
685 /* Propogate the count up the 'totals' tree */
686 instance->counters.ce_count++;
687 edac_dev->counters.ce_count++;
689 if (edac_device_get_log_ce(edac_dev))
690 edac_device_printk(edac_dev, KERN_WARNING,
691 "CE: %s instance: %s block: %s '%s'\n",
692 edac_dev->ctl_name, instance->name,
693 block ? block->name : "N/A", msg);
695 EXPORT_SYMBOL_GPL(edac_device_handle_ce);
698 * edac_device_handle_ue
699 * perform a common output and handling of an 'edac_dev' UE event
701 void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
702 int inst_nr, int block_nr, const char *msg)
704 struct edac_device_instance *instance;
705 struct edac_device_block *block = NULL;
707 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
708 edac_device_printk(edac_dev, KERN_ERR,
709 "INTERNAL ERROR: 'instance' out of range "
710 "(%d >= %d)\n", inst_nr,
711 edac_dev->nr_instances);
715 instance = edac_dev->instances + inst_nr;
717 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
718 edac_device_printk(edac_dev, KERN_ERR,
719 "INTERNAL ERROR: instance %d 'block' "
720 "out of range (%d >= %d)\n",
722 instance->nr_blocks);
726 if (instance->nr_blocks > 0) {
727 block = instance->blocks + block_nr;
728 block->counters.ue_count++;
731 /* Propogate the count up the 'totals' tree */
732 instance->counters.ue_count++;
733 edac_dev->counters.ue_count++;
735 if (edac_device_get_log_ue(edac_dev))
736 edac_device_printk(edac_dev, KERN_EMERG,
737 "UE: %s instance: %s block: %s '%s'\n",
738 edac_dev->ctl_name, instance->name,
739 block ? block->name : "N/A", msg);
741 if (edac_device_get_panic_on_ue(edac_dev))
742 panic("EDAC %s: UE instance: %s block %s '%s'\n",
743 edac_dev->ctl_name, instance->name,
744 block ? block->name : "N/A", msg);
746 EXPORT_SYMBOL_GPL(edac_device_handle_ue);