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 to memory controller's control array 'edac_device_list' */
36 static DEFINE_MUTEX(device_ctls_mutex);
37 static struct list_head edac_device_list = LIST_HEAD_INIT(edac_device_list);
39 #ifdef CONFIG_EDAC_DEBUG
40 static void edac_device_dump_device(struct edac_device_ctl_info *edac_dev)
42 debugf3("\tedac_dev = %p dev_idx=%d \n", edac_dev, edac_dev->dev_idx);
43 debugf4("\tedac_dev->edac_check = %p\n", edac_dev->edac_check);
44 debugf3("\tdev = %p\n", edac_dev->dev);
45 debugf3("\tmod_name:ctl_name = %s:%s\n",
46 edac_dev->mod_name, edac_dev->ctl_name);
47 debugf3("\tpvt_info = %p\n\n", edac_dev->pvt_info);
49 #endif /* CONFIG_EDAC_DEBUG */
52 * edac_device_alloc_ctl_info()
53 * Allocate a new edac device control info structure
55 * The control structure is allocated in complete chunk
56 * from the OS. It is in turn sub allocated to the
57 * various objects that compose the struture
59 * The structure has a 'nr_instance' array within itself.
60 * Each instance represents a major component
61 * Example: L1 cache and L2 cache are 2 instance components
63 * Within each instance is an array of 'nr_blocks' blockoffsets
65 struct edac_device_ctl_info *edac_device_alloc_ctl_info(
67 char *edac_device_name, unsigned nr_instances,
68 char *edac_block_name, unsigned nr_blocks,
69 unsigned offset_value, /* zero, 1, or other based offset */
70 struct edac_dev_sysfs_block_attribute *attrib_spec, unsigned nr_attrib,
73 struct edac_device_ctl_info *dev_ctl;
74 struct edac_device_instance *dev_inst, *inst;
75 struct edac_device_block *dev_blk, *blk_p, *blk;
76 struct edac_dev_sysfs_block_attribute *dev_attrib, *attrib_p, *attrib;
79 unsigned instance, block, attr;
82 debugf1("%s() instances=%d blocks=%d\n",
83 __func__, nr_instances, nr_blocks);
85 /* Calculate the size of memory we need to allocate AND
86 * determine the offsets of the various item arrays
87 * (instance,block,attrib) from the start of an allocated structure.
88 * We want the alignment of each item (instance,block,attrib)
89 * to be at least as stringent as what the compiler would
90 * provide if we could simply hardcode everything into a single struct.
92 dev_ctl = (struct edac_device_ctl_info *)NULL;
94 /* Calc the 'end' offset past end of ONE ctl_info structure
95 * which will become the start of the 'instance' array
97 dev_inst = edac_align_ptr(&dev_ctl[1], sizeof(*dev_inst));
99 /* Calc the 'end' offset past the instance array within the ctl_info
100 * which will become the start of the block array
102 dev_blk = edac_align_ptr(&dev_inst[nr_instances], sizeof(*dev_blk));
104 /* Calc the 'end' offset past the dev_blk array
105 * which will become the start of the attrib array, if any.
107 count = nr_instances * nr_blocks;
108 dev_attrib = edac_align_ptr(&dev_blk[count], sizeof(*dev_attrib));
110 /* Check for case of when an attribute array is specified */
112 /* calc how many nr_attrib we need */
115 /* Calc the 'end' offset past the attributes array */
116 pvt = edac_align_ptr(&dev_attrib[count], sz_private);
118 /* no attribute array specificed */
119 pvt = edac_align_ptr(dev_attrib, sz_private);
122 /* 'pvt' now points to where the private data area is.
123 * At this point 'pvt' (like dev_inst,dev_blk and dev_attrib)
124 * is baselined at ZERO
126 total_size = ((unsigned long)pvt) + sz_private;
128 /* Allocate the amount of memory for the set of control structures */
129 dev_ctl = kzalloc(total_size, GFP_KERNEL);
133 /* Adjust pointers so they point within the actual memory we
134 * just allocated rather than an imaginary chunk of memory
135 * located at address 0.
136 * 'dev_ctl' points to REAL memory, while the others are
137 * ZERO based and thus need to be adjusted to point within
138 * the allocated memory.
140 dev_inst = (struct edac_device_instance *)
141 (((char *)dev_ctl) + ((unsigned long)dev_inst));
142 dev_blk = (struct edac_device_block *)
143 (((char *)dev_ctl) + ((unsigned long)dev_blk));
144 dev_attrib = (struct edac_dev_sysfs_block_attribute *)
145 (((char *)dev_ctl) + ((unsigned long)dev_attrib));
146 pvt = sz_private ? (((char *)dev_ctl) + ((unsigned long)pvt)) : NULL;
148 /* Begin storing the information into the control info structure */
149 dev_ctl->dev_idx = device_index;
150 dev_ctl->nr_instances = nr_instances;
151 dev_ctl->instances = dev_inst;
152 dev_ctl->pvt_info = pvt;
154 /* Name of this edac device */
155 snprintf(dev_ctl->name,sizeof(dev_ctl->name),"%s",edac_device_name);
157 /* Initialize every Instance */
158 for (instance = 0; instance < nr_instances; instance++) {
159 inst = &dev_inst[instance];
161 inst->nr_blocks = nr_blocks;
162 blk_p = &dev_blk[instance * nr_blocks];
163 inst->blocks = blk_p;
165 /* name of this instance */
166 snprintf(inst->name, sizeof(inst->name),
167 "%s%u", edac_device_name, instance);
169 /* Initialize every block in each instance */
170 for (block = 0; block < nr_blocks; block++) {
172 blk->instance = inst;
173 snprintf(blk->name, sizeof(blk->name),
174 "%s%d", edac_block_name, block+offset_value);
176 debugf1("%s() instance=%d block=%d name=%s\n",
177 __func__, instance, block, blk->name);
179 /* if there are NO attributes OR no attribute pointer
180 * then continue on to next block iteration
182 if ((nr_attrib == 0) || (attrib_spec == NULL))
185 /* setup the attribute array for this block */
186 blk->nr_attribs = nr_attrib;
187 attrib_p = &dev_attrib[block*nr_instances*nr_attrib];
188 blk->block_attributes = attrib_p;
190 /* Initialize every user specified attribute in this
191 * block with the data the caller passed in
193 for (attr = 0; attr < nr_attrib; attr++) {
194 attrib = &attrib_p[attr];
195 attrib->attr = attrib_spec->attr;
196 attrib->show = attrib_spec->show;
197 attrib->store = attrib_spec->store;
199 /* up reference this block */
202 /* bump the attrib_spec */
208 /* Mark this instance as merely ALLOCATED */
209 dev_ctl->op_state = OP_ALLOC;
213 EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
216 * edac_device_free_ctl_info()
217 * frees the memory allocated by the edac_device_alloc_ctl_info()
220 void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)
224 EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
227 * find_edac_device_by_dev
228 * scans the edac_device list for a specific 'struct device *'
230 * lock to be held prior to call: device_ctls_mutex
233 * pointer to control structure managing 'dev'
234 * NULL if not found on list
236 static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
238 struct edac_device_ctl_info *edac_dev;
239 struct list_head *item;
241 debugf3("%s()\n", __func__);
243 list_for_each(item, &edac_device_list) {
244 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
246 if (edac_dev->dev == dev)
254 * add_edac_dev_to_global_list
255 * Before calling this function, caller must
256 * assign a unique value to edac_dev->dev_idx.
258 * lock to be held prior to call: device_ctls_mutex
264 static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
266 struct list_head *item, *insert_before;
267 struct edac_device_ctl_info *rover;
269 insert_before = &edac_device_list;
271 /* Determine if already on the list */
272 rover = find_edac_device_by_dev(edac_dev->dev);
273 if (unlikely(rover != NULL))
276 /* Insert in ascending order by 'dev_idx', so find position */
277 list_for_each(item, &edac_device_list) {
278 rover = list_entry(item, struct edac_device_ctl_info, link);
280 if (rover->dev_idx >= edac_dev->dev_idx) {
281 if (unlikely(rover->dev_idx == edac_dev->dev_idx))
284 insert_before = item;
289 list_add_tail_rcu(&edac_dev->link, insert_before);
293 edac_printk(KERN_WARNING, EDAC_MC,
294 "%s (%s) %s %s already assigned %d\n",
295 rover->dev->bus_id, dev_name(rover),
296 rover->mod_name, rover->ctl_name, rover->dev_idx);
300 edac_printk(KERN_WARNING, EDAC_MC,
301 "bug in low-level driver: attempt to assign\n"
302 " duplicate dev_idx %d in %s()\n", rover->dev_idx,
308 * complete_edac_device_list_del
310 * callback function when reference count is zero
312 static void complete_edac_device_list_del(struct rcu_head *head)
314 struct edac_device_ctl_info *edac_dev;
316 edac_dev = container_of(head, struct edac_device_ctl_info, rcu);
317 INIT_LIST_HEAD(&edac_dev->link);
318 complete(&edac_dev->complete);
322 * del_edac_device_from_global_list
324 * remove the RCU, setup for a callback call, then wait for the
327 static void del_edac_device_from_global_list(struct edac_device_ctl_info
330 list_del_rcu(&edac_device->link);
331 init_completion(&edac_device->complete);
332 call_rcu(&edac_device->rcu, complete_edac_device_list_del);
333 wait_for_completion(&edac_device->complete);
338 * Search for a edac_device_ctl_info structure whose index is 'idx'.
340 * If found, return a pointer to the structure.
343 * Caller must hold device_ctls_mutex.
345 struct edac_device_ctl_info *edac_device_find(int idx)
347 struct list_head *item;
348 struct edac_device_ctl_info *edac_dev;
350 /* Iterate over list, looking for exact match of ID */
351 list_for_each(item, &edac_device_list) {
352 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
354 if (edac_dev->dev_idx >= idx) {
355 if (edac_dev->dev_idx == idx)
358 /* not on list, so terminate early */
365 EXPORT_SYMBOL_GPL(edac_device_find);
368 * edac_device_workq_function
369 * performs the operation scheduled by a workq request
371 static void edac_device_workq_function(struct work_struct *work_req)
373 struct delayed_work *d_work = (struct delayed_work *)work_req;
374 struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);
376 //debugf0("%s() here and running\n", __func__);
377 mutex_lock(&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);
388 queue_delayed_work(edac_workqueue, &edac_dev->work, edac_dev->delay);
392 * edac_device_workq_setup
393 * initialize a workq item for this edac_device instance
394 * passing in the new delay period in msec
396 void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
399 debugf0("%s()\n", __func__);
401 edac_dev->poll_msec = msec;
402 edac_dev->delay = msecs_to_jiffies(msec); /* Calc delay jiffies */
404 INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
405 queue_delayed_work(edac_workqueue, &edac_dev->work, edac_dev->delay);
409 * edac_device_workq_teardown
410 * stop the workq processing on this edac_dev
412 void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)
416 status = cancel_delayed_work(&edac_dev->work);
418 /* workq instance might be running, wait for it */
419 flush_workqueue(edac_workqueue);
424 * edac_device_reset_delay_period
427 void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,
430 mutex_lock(&device_ctls_mutex);
432 /* cancel the current workq request */
433 edac_device_workq_teardown(edac_dev);
435 /* restart the workq request, with new delay value */
436 edac_device_workq_setup(edac_dev, value);
438 mutex_unlock(&device_ctls_mutex);
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_device' structure.
452 int edac_device_add_device(struct edac_device_ctl_info *edac_dev)
454 debugf0("%s()\n", __func__);
456 #ifdef CONFIG_EDAC_DEBUG
457 if (edac_debug_level >= 3)
458 edac_device_dump_device(edac_dev);
460 mutex_lock(&device_ctls_mutex);
462 if (add_edac_dev_to_global_list(edac_dev))
465 /* set load time so that error rate can be tracked */
466 edac_dev->start_time = jiffies;
468 /* create this instance's sysfs entries */
469 if (edac_device_create_sysfs(edac_dev)) {
470 edac_device_printk(edac_dev, KERN_WARNING,
471 "failed to create sysfs device\n");
475 /* If there IS a check routine, then we are running POLLED */
476 if (edac_dev->edac_check != NULL) {
477 /* This instance is NOW RUNNING */
478 edac_dev->op_state = OP_RUNNING_POLL;
481 * enable workq processing on this instance,
482 * default = 1000 msec
484 edac_device_workq_setup(edac_dev, 1000);
486 edac_dev->op_state = OP_RUNNING_INTERRUPT;
489 /* Report action taken */
490 edac_device_printk(edac_dev, KERN_INFO,
491 "Giving out device to module '%s' controller "
492 "'%s': DEV '%s' (%s)\n",
496 edac_op_state_to_string(edac_dev->op_state));
498 mutex_unlock(&device_ctls_mutex);
502 /* Some error, so remove the entry from the lsit */
503 del_edac_device_from_global_list(edac_dev);
506 mutex_unlock(&device_ctls_mutex);
509 EXPORT_SYMBOL_GPL(edac_device_add_device);
512 * edac_device_del_device:
513 * Remove sysfs entries for specified edac_device structure and
514 * then remove edac_device structure from global list
517 * Pointer to 'struct device' representing edac_device
518 * structure to remove.
521 * Pointer to removed edac_device structure,
522 * OR NULL if device not found.
524 struct edac_device_ctl_info *edac_device_del_device(struct device *dev)
526 struct edac_device_ctl_info *edac_dev;
528 debugf0("MC: %s()\n", __func__);
530 mutex_lock(&device_ctls_mutex);
532 /* Find the structure on the list, if not there, then leave */
533 edac_dev = find_edac_device_by_dev(dev);
534 if (edac_dev == NULL) {
535 mutex_unlock(&device_ctls_mutex);
539 /* mark this instance as OFFLINE */
540 edac_dev->op_state = OP_OFFLINE;
542 /* clear workq processing on this instance */
543 edac_device_workq_teardown(edac_dev);
545 /* Tear down the sysfs entries for this instance */
546 edac_device_remove_sysfs(edac_dev);
548 /* deregister from global list */
549 del_edac_device_from_global_list(edac_dev);
551 mutex_unlock(&device_ctls_mutex);
553 edac_printk(KERN_INFO, EDAC_MC,
554 "Removed device %d for %s %s: DEV %s\n",
556 edac_dev->mod_name, edac_dev->ctl_name, dev_name(edac_dev));
560 EXPORT_SYMBOL_GPL(edac_device_del_device);
562 static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
564 return edac_dev->log_ce;
567 static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
569 return edac_dev->log_ue;
572 static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
575 return edac_dev->panic_on_ue;
579 * edac_device_handle_ce
580 * perform a common output and handling of an 'edac_dev' CE event
582 void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
583 int inst_nr, int block_nr, const char *msg)
585 struct edac_device_instance *instance;
586 struct edac_device_block *block = NULL;
588 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
589 edac_device_printk(edac_dev, KERN_ERR,
590 "INTERNAL ERROR: 'instance' out of range "
591 "(%d >= %d)\n", inst_nr,
592 edac_dev->nr_instances);
596 instance = edac_dev->instances + inst_nr;
598 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
599 edac_device_printk(edac_dev, KERN_ERR,
600 "INTERNAL ERROR: instance %d 'block' "
601 "out of range (%d >= %d)\n",
603 instance->nr_blocks);
607 if (instance->nr_blocks > 0) {
608 block = instance->blocks + block_nr;
609 block->counters.ce_count++;
612 /* Propogate the count up the 'totals' tree */
613 instance->counters.ce_count++;
614 edac_dev->counters.ce_count++;
616 if (edac_device_get_log_ce(edac_dev))
617 edac_device_printk(edac_dev, KERN_WARNING,
618 "CE: %s instance: %s block: %s '%s'\n",
619 edac_dev->ctl_name, instance->name,
620 block ? block->name : "N/A", msg);
622 EXPORT_SYMBOL_GPL(edac_device_handle_ce);
625 * edac_device_handle_ue
626 * perform a common output and handling of an 'edac_dev' UE event
628 void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
629 int inst_nr, int block_nr, const char *msg)
631 struct edac_device_instance *instance;
632 struct edac_device_block *block = NULL;
634 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
635 edac_device_printk(edac_dev, KERN_ERR,
636 "INTERNAL ERROR: 'instance' out of range "
637 "(%d >= %d)\n", inst_nr,
638 edac_dev->nr_instances);
642 instance = edac_dev->instances + inst_nr;
644 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
645 edac_device_printk(edac_dev, KERN_ERR,
646 "INTERNAL ERROR: instance %d 'block' "
647 "out of range (%d >= %d)\n",
649 instance->nr_blocks);
653 if (instance->nr_blocks > 0) {
654 block = instance->blocks + block_nr;
655 block->counters.ue_count++;
658 /* Propogate the count up the 'totals' tree */
659 instance->counters.ue_count++;
660 edac_dev->counters.ue_count++;
662 if (edac_device_get_log_ue(edac_dev))
663 edac_device_printk(edac_dev, KERN_EMERG,
664 "UE: %s instance: %s block: %s '%s'\n",
665 edac_dev->ctl_name, instance->name,
666 block ? block->name : "N/A", msg);
668 if (edac_device_get_panic_on_ue(edac_dev))
669 panic("EDAC %s: UE instance: %s block %s '%s'\n",
670 edac_dev->ctl_name, instance->name,
671 block ? block->name : "N/A", msg);
673 EXPORT_SYMBOL_GPL(edac_device_handle_ue);