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_attrib_spec *attrib_spec, unsigned nr_attribs)
72 struct edac_device_ctl_info *dev_ctl;
73 struct edac_device_instance *dev_inst, *inst;
74 struct edac_device_block *dev_blk, *blk_p, *blk;
75 struct edac_attrib *dev_attrib, *attrib_p, *attrib;
78 unsigned instance, block, attr;
81 debugf1("%s() instances=%d blocks=%d\n",
82 __func__, nr_instances, nr_blocks);
84 /* Figure out the offsets of the various items from the start of an
85 * ctl_info structure. We want the alignment of each item
86 * to be at least as stringent as what the compiler would
87 * provide if we could simply hardcode everything into a single struct.
89 dev_ctl = (struct edac_device_ctl_info *)NULL;
91 /* Calc the 'end' offset past the ctl_info structure */
92 dev_inst = edac_align_ptr(&dev_ctl[1], sizeof(*dev_inst));
94 /* Calc the 'end' offset past the instance array */
95 dev_blk = edac_align_ptr(&dev_inst[nr_instances], sizeof(*dev_blk));
97 /* Calc the 'end' offset past the dev_blk array */
98 count = nr_instances * nr_blocks;
99 dev_attrib = edac_align_ptr(&dev_blk[count], sizeof(*dev_attrib));
101 /* Check for case of NO attributes specified */
105 /* Calc the 'end' offset past the attributes array */
106 pvt = edac_align_ptr(&dev_attrib[count], sz_private);
107 total_size = ((unsigned long)pvt) + sz_private;
109 /* Allocate the amount of memory for the set of control structures */
110 dev_ctl = kzalloc(total_size, GFP_KERNEL);
114 /* Adjust pointers so they point within the memory we just allocated
115 * rather than an imaginary chunk of memory located at address 0.
117 dev_inst = (struct edac_device_instance *)
118 (((char *)dev_ctl) + ((unsigned long)dev_inst));
119 dev_blk = (struct edac_device_block *)
120 (((char *)dev_ctl) + ((unsigned long)dev_blk));
121 dev_attrib = (struct edac_attrib *)
122 (((char *)dev_ctl) + ((unsigned long)dev_attrib));
123 pvt = sz_private ? (((char *)dev_ctl) + ((unsigned long)pvt)) : NULL;
125 dev_ctl->nr_instances = nr_instances;
126 dev_ctl->instances = dev_inst;
127 dev_ctl->pvt_info = pvt;
129 /* Name of this edac device */
130 snprintf(dev_ctl->name,sizeof(dev_ctl->name),"%s",edac_device_name);
132 /* Initialize every Instance */
133 for (instance = 0; instance < nr_instances; instance++) {
134 inst = &dev_inst[instance];
136 inst->nr_blocks = nr_blocks;
137 blk_p = &dev_blk[instance * nr_blocks];
138 inst->blocks = blk_p;
140 /* name of this instance */
141 snprintf(inst->name, sizeof(inst->name),
142 "%s%u", edac_device_name, instance);
144 /* Initialize every block in each instance */
145 for (block = 0; block < nr_blocks; block++) {
147 blk->instance = inst;
148 blk->nr_attribs = nr_attribs;
149 attrib_p = &dev_attrib[block * nr_attribs];
150 blk->attribs = attrib_p;
151 snprintf(blk->name, sizeof(blk->name),
152 "%s%d", edac_block_name, block+offset_value);
154 debugf1("%s() instance=%d block=%d name=%s\n",
155 __func__, instance, block, blk->name);
157 if (attrib_spec != NULL) {
158 /* when there is an attrib_spec passed int then
159 * Initialize every attrib of each block
161 for (attr = 0; attr < nr_attribs; attr++) {
162 attrib = &attrib_p[attr];
165 /* Link each attribute to the caller's
166 * spec entry, for name and type
168 attrib->spec = &attrib_spec[attr];
174 /* Mark this instance as merely ALLOCATED */
175 dev_ctl->op_state = OP_ALLOC;
179 EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info);
182 * edac_device_free_ctl_info()
183 * frees the memory allocated by the edac_device_alloc_ctl_info()
186 void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info)
190 EXPORT_SYMBOL_GPL(edac_device_free_ctl_info);
193 * find_edac_device_by_dev
194 * scans the edac_device list for a specific 'struct device *'
196 * lock to be held prior to call: device_ctls_mutex
199 * pointer to control structure managing 'dev'
200 * NULL if not found on list
202 static struct edac_device_ctl_info *find_edac_device_by_dev(struct device *dev)
204 struct edac_device_ctl_info *edac_dev;
205 struct list_head *item;
207 debugf3("%s()\n", __func__);
209 list_for_each(item, &edac_device_list) {
210 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
212 if (edac_dev->dev == dev)
220 * add_edac_dev_to_global_list
221 * Before calling this function, caller must
222 * assign a unique value to edac_dev->dev_idx.
224 * lock to be held prior to call: device_ctls_mutex
230 static int add_edac_dev_to_global_list(struct edac_device_ctl_info *edac_dev)
232 struct list_head *item, *insert_before;
233 struct edac_device_ctl_info *rover;
235 insert_before = &edac_device_list;
237 /* Determine if already on the list */
238 rover = find_edac_device_by_dev(edac_dev->dev);
239 if (unlikely(rover != NULL))
242 /* Insert in ascending order by 'dev_idx', so find position */
243 list_for_each(item, &edac_device_list) {
244 rover = list_entry(item, struct edac_device_ctl_info, link);
246 if (rover->dev_idx >= edac_dev->dev_idx) {
247 if (unlikely(rover->dev_idx == edac_dev->dev_idx))
250 insert_before = item;
255 list_add_tail_rcu(&edac_dev->link, insert_before);
259 edac_printk(KERN_WARNING, EDAC_MC,
260 "%s (%s) %s %s already assigned %d\n",
261 rover->dev->bus_id, dev_name(rover),
262 rover->mod_name, rover->ctl_name, rover->dev_idx);
266 edac_printk(KERN_WARNING, EDAC_MC,
267 "bug in low-level driver: attempt to assign\n"
268 " duplicate dev_idx %d in %s()\n", rover->dev_idx,
274 * complete_edac_device_list_del
276 * callback function when reference count is zero
278 static void complete_edac_device_list_del(struct rcu_head *head)
280 struct edac_device_ctl_info *edac_dev;
282 edac_dev = container_of(head, struct edac_device_ctl_info, rcu);
283 INIT_LIST_HEAD(&edac_dev->link);
284 complete(&edac_dev->complete);
288 * del_edac_device_from_global_list
290 * remove the RCU, setup for a callback call, then wait for the
293 static void del_edac_device_from_global_list(struct edac_device_ctl_info
296 list_del_rcu(&edac_device->link);
297 init_completion(&edac_device->complete);
298 call_rcu(&edac_device->rcu, complete_edac_device_list_del);
299 wait_for_completion(&edac_device->complete);
304 * Search for a edac_device_ctl_info structure whose index is 'idx'.
306 * If found, return a pointer to the structure.
309 * Caller must hold device_ctls_mutex.
311 struct edac_device_ctl_info *edac_device_find(int idx)
313 struct list_head *item;
314 struct edac_device_ctl_info *edac_dev;
316 /* Iterate over list, looking for exact match of ID */
317 list_for_each(item, &edac_device_list) {
318 edac_dev = list_entry(item, struct edac_device_ctl_info, link);
320 if (edac_dev->dev_idx >= idx) {
321 if (edac_dev->dev_idx == idx)
324 /* not on list, so terminate early */
331 EXPORT_SYMBOL_GPL(edac_device_find);
334 * edac_device_workq_function
335 * performs the operation scheduled by a workq request
337 static void edac_device_workq_function(struct work_struct *work_req)
339 struct delayed_work *d_work = (struct delayed_work *)work_req;
340 struct edac_device_ctl_info *edac_dev = to_edac_device_ctl_work(d_work);
342 //debugf0("%s() here and running\n", __func__);
343 mutex_lock(&device_ctls_mutex);
345 /* Only poll controllers that are running polled and have a check */
346 if ((edac_dev->op_state == OP_RUNNING_POLL) &&
347 (edac_dev->edac_check != NULL)) {
348 edac_dev->edac_check(edac_dev);
351 mutex_unlock(&device_ctls_mutex);
354 queue_delayed_work(edac_workqueue, &edac_dev->work, edac_dev->delay);
358 * edac_device_workq_setup
359 * initialize a workq item for this edac_device instance
360 * passing in the new delay period in msec
362 void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
365 debugf0("%s()\n", __func__);
367 edac_dev->poll_msec = msec;
368 edac_dev->delay = msecs_to_jiffies(msec); /* Calc delay jiffies */
370 INIT_DELAYED_WORK(&edac_dev->work, edac_device_workq_function);
371 queue_delayed_work(edac_workqueue, &edac_dev->work, edac_dev->delay);
375 * edac_device_workq_teardown
376 * stop the workq processing on this edac_dev
378 void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev)
382 status = cancel_delayed_work(&edac_dev->work);
384 /* workq instance might be running, wait for it */
385 flush_workqueue(edac_workqueue);
390 * edac_device_reset_delay_period
393 void edac_device_reset_delay_period(struct edac_device_ctl_info *edac_dev,
396 mutex_lock(&device_ctls_mutex);
398 /* cancel the current workq request */
399 edac_device_workq_teardown(edac_dev);
401 /* restart the workq request, with new delay value */
402 edac_device_workq_setup(edac_dev, value);
404 mutex_unlock(&device_ctls_mutex);
408 * edac_device_add_device: Insert the 'edac_dev' structure into the
409 * edac_device global list and create sysfs entries associated with
410 * edac_device structure.
411 * @edac_device: pointer to the edac_device structure to be added to the list
412 * @edac_idx: A unique numeric identifier to be assigned to the
413 * 'edac_device' structure.
419 int edac_device_add_device(struct edac_device_ctl_info *edac_dev, int edac_idx)
421 debugf0("%s()\n", __func__);
423 edac_dev->dev_idx = edac_idx;
424 #ifdef CONFIG_EDAC_DEBUG
425 if (edac_debug_level >= 3)
426 edac_device_dump_device(edac_dev);
428 mutex_lock(&device_ctls_mutex);
430 if (add_edac_dev_to_global_list(edac_dev))
433 /* set load time so that error rate can be tracked */
434 edac_dev->start_time = jiffies;
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");
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;
449 * enable workq processing on this instance,
450 * default = 1000 msec
452 edac_device_workq_setup(edac_dev, 1000);
454 edac_dev->op_state = OP_RUNNING_INTERRUPT;
457 /* Report action taken */
458 edac_device_printk(edac_dev, KERN_INFO,
459 "Giving out device to module '%s' controller "
460 "'%s': DEV '%s' (%s)\n",
464 edac_op_state_to_string(edac_dev->op_state));
466 mutex_unlock(&device_ctls_mutex);
470 /* Some error, so remove the entry from the lsit */
471 del_edac_device_from_global_list(edac_dev);
474 mutex_unlock(&device_ctls_mutex);
477 EXPORT_SYMBOL_GPL(edac_device_add_device);
480 * edac_device_del_device:
481 * Remove sysfs entries for specified edac_device structure and
482 * then remove edac_device structure from global list
485 * Pointer to 'struct device' representing edac_device
486 * structure to remove.
489 * Pointer to removed edac_device structure,
490 * OR NULL if device not found.
492 struct edac_device_ctl_info *edac_device_del_device(struct device *dev)
494 struct edac_device_ctl_info *edac_dev;
496 debugf0("MC: %s()\n", __func__);
498 mutex_lock(&device_ctls_mutex);
500 /* Find the structure on the list, if not there, then leave */
501 edac_dev = find_edac_device_by_dev(dev);
502 if (edac_dev == NULL) {
503 mutex_unlock(&device_ctls_mutex);
507 /* mark this instance as OFFLINE */
508 edac_dev->op_state = OP_OFFLINE;
510 /* clear workq processing on this instance */
511 edac_device_workq_teardown(edac_dev);
513 /* Tear down the sysfs entries for this instance */
514 edac_device_remove_sysfs(edac_dev);
516 /* deregister from global list */
517 del_edac_device_from_global_list(edac_dev);
519 mutex_unlock(&device_ctls_mutex);
521 edac_printk(KERN_INFO, EDAC_MC,
522 "Removed device %d for %s %s: DEV %s\n",
524 edac_dev->mod_name, edac_dev->ctl_name, dev_name(edac_dev));
528 EXPORT_SYMBOL_GPL(edac_device_del_device);
530 static inline int edac_device_get_log_ce(struct edac_device_ctl_info *edac_dev)
532 return edac_dev->log_ce;
535 static inline int edac_device_get_log_ue(struct edac_device_ctl_info *edac_dev)
537 return edac_dev->log_ue;
540 static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
543 return edac_dev->panic_on_ue;
547 * edac_device_handle_ce
548 * perform a common output and handling of an 'edac_dev' CE event
550 void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
551 int inst_nr, int block_nr, const char *msg)
553 struct edac_device_instance *instance;
554 struct edac_device_block *block = NULL;
556 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
557 edac_device_printk(edac_dev, KERN_ERR,
558 "INTERNAL ERROR: 'instance' out of range "
559 "(%d >= %d)\n", inst_nr,
560 edac_dev->nr_instances);
564 instance = edac_dev->instances + inst_nr;
566 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
567 edac_device_printk(edac_dev, KERN_ERR,
568 "INTERNAL ERROR: instance %d 'block' "
569 "out of range (%d >= %d)\n",
571 instance->nr_blocks);
575 if (instance->nr_blocks > 0) {
576 block = instance->blocks + block_nr;
577 block->counters.ce_count++;
580 /* Propogate the count up the 'totals' tree */
581 instance->counters.ce_count++;
582 edac_dev->counters.ce_count++;
584 if (edac_device_get_log_ce(edac_dev))
585 edac_device_printk(edac_dev, KERN_WARNING,
586 "CE: %s instance: %s block: %s '%s'\n",
587 edac_dev->ctl_name, instance->name,
588 block ? block->name : "N/A", msg);
590 EXPORT_SYMBOL_GPL(edac_device_handle_ce);
593 * edac_device_handle_ue
594 * perform a common output and handling of an 'edac_dev' UE event
596 void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
597 int inst_nr, int block_nr, const char *msg)
599 struct edac_device_instance *instance;
600 struct edac_device_block *block = NULL;
602 if ((inst_nr >= edac_dev->nr_instances) || (inst_nr < 0)) {
603 edac_device_printk(edac_dev, KERN_ERR,
604 "INTERNAL ERROR: 'instance' out of range "
605 "(%d >= %d)\n", inst_nr,
606 edac_dev->nr_instances);
610 instance = edac_dev->instances + inst_nr;
612 if ((block_nr >= instance->nr_blocks) || (block_nr < 0)) {
613 edac_device_printk(edac_dev, KERN_ERR,
614 "INTERNAL ERROR: instance %d 'block' "
615 "out of range (%d >= %d)\n",
617 instance->nr_blocks);
621 if (instance->nr_blocks > 0) {
622 block = instance->blocks + block_nr;
623 block->counters.ue_count++;
626 /* Propogate the count up the 'totals' tree */
627 instance->counters.ue_count++;
628 edac_dev->counters.ue_count++;
630 if (edac_device_get_log_ue(edac_dev))
631 edac_device_printk(edac_dev, KERN_EMERG,
632 "UE: %s instance: %s block: %s '%s'\n",
633 edac_dev->ctl_name, instance->name,
634 block ? block->name : "N/A", msg);
636 if (edac_device_get_panic_on_ue(edac_dev))
637 panic("EDAC %s: UE instance: %s block %s '%s'\n",
638 edac_dev->ctl_name, instance->name,
639 block ? block->name : "N/A", msg);
641 EXPORT_SYMBOL_GPL(edac_device_handle_ue);