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 LIST_HEAD(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 /* Default logging of CEs and UEs */
162 /* Name of this edac device */
163 snprintf(dev_ctl
->name
,sizeof(dev_ctl
->name
),"%s",edac_device_name
);
165 debugf4("%s() edac_dev=%p next after end=%p\n",
166 __func__
, dev_ctl
, pvt
+ sz_private
);
168 /* Initialize every Instance */
169 for (instance
= 0; instance
< nr_instances
; instance
++) {
170 inst
= &dev_inst
[instance
];
172 inst
->nr_blocks
= nr_blocks
;
173 blk_p
= &dev_blk
[instance
* nr_blocks
];
174 inst
->blocks
= blk_p
;
176 /* name of this instance */
177 snprintf(inst
->name
, sizeof(inst
->name
),
178 "%s%u", edac_device_name
, instance
);
180 /* Initialize every block in each instance */
181 for (block
= 0; block
< nr_blocks
; block
++) {
183 blk
->instance
= inst
;
184 snprintf(blk
->name
, sizeof(blk
->name
),
185 "%s%d", edac_block_name
, block
+offset_value
);
187 debugf4("%s() instance=%d inst_p=%p block=#%d "
188 "block_p=%p name='%s'\n",
189 __func__
, instance
, inst
, block
,
192 /* if there are NO attributes OR no attribute pointer
193 * then continue on to next block iteration
195 if ((nr_attrib
== 0) || (attrib_spec
== NULL
))
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
;
203 debugf4("%s() THIS BLOCK_ATTRIB=%p\n",
204 __func__
, blk
->block_attributes
);
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'
211 for (attr
= 0; attr
< nr_attrib
; attr
++) {
212 attrib
= &attrib_p
[attr
];
214 /* populate the unique per attrib
215 * with the code pointers and info
217 attrib
->attr
= attrib_spec
[attr
].attr
;
218 attrib
->show
= attrib_spec
[attr
].show
;
219 attrib
->store
= attrib_spec
[attr
].store
;
221 attrib
->block
= blk
; /* up link */
223 debugf4("%s() alloc-attrib=%p attrib_name='%s' "
224 "attrib-spec=%p spec-name=%s\n",
225 __func__
, attrib
, attrib
->attr
.name
,
227 attrib_spec
[attr
].attr
.name
233 /* Mark this instance as merely ALLOCATED */
234 dev_ctl
->op_state
= OP_ALLOC
;
237 * Initialize the 'root' kobj for the edac_device controller
239 err
= edac_device_register_sysfs_main_kobj(dev_ctl
);
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
254 EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info
);
257 * edac_device_free_ctl_info()
258 * frees the memory allocated by the edac_device_alloc_ctl_info()
261 void edac_device_free_ctl_info(struct edac_device_ctl_info
*ctl_info
)
263 edac_device_unregister_sysfs_main_kobj(ctl_info
);
265 EXPORT_SYMBOL_GPL(edac_device_free_ctl_info
);
268 * find_edac_device_by_dev
269 * scans the edac_device list for a specific 'struct device *'
271 * lock to be held prior to call: device_ctls_mutex
274 * pointer to control structure managing 'dev'
275 * NULL if not found on list
277 static struct edac_device_ctl_info
*find_edac_device_by_dev(struct device
*dev
)
279 struct edac_device_ctl_info
*edac_dev
;
280 struct list_head
*item
;
282 debugf0("%s()\n", __func__
);
284 list_for_each(item
, &edac_device_list
) {
285 edac_dev
= list_entry(item
, struct edac_device_ctl_info
, link
);
287 if (edac_dev
->dev
== dev
)
295 * add_edac_dev_to_global_list
296 * Before calling this function, caller must
297 * assign a unique value to edac_dev->dev_idx.
299 * lock to be held prior to call: device_ctls_mutex
305 static int add_edac_dev_to_global_list(struct edac_device_ctl_info
*edac_dev
)
307 struct list_head
*item
, *insert_before
;
308 struct edac_device_ctl_info
*rover
;
310 insert_before
= &edac_device_list
;
312 /* Determine if already on the list */
313 rover
= find_edac_device_by_dev(edac_dev
->dev
);
314 if (unlikely(rover
!= NULL
))
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
);
321 if (rover
->dev_idx
>= edac_dev
->dev_idx
) {
322 if (unlikely(rover
->dev_idx
== edac_dev
->dev_idx
))
325 insert_before
= item
;
330 list_add_tail_rcu(&edac_dev
->link
, insert_before
);
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
);
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
,
349 * complete_edac_device_list_del
351 * callback function when reference count is zero
353 static void complete_edac_device_list_del(struct rcu_head
*head
)
355 struct edac_device_ctl_info
*edac_dev
;
357 edac_dev
= container_of(head
, struct edac_device_ctl_info
, rcu
);
358 INIT_LIST_HEAD(&edac_dev
->link
);
359 complete(&edac_dev
->removal_complete
);
363 * del_edac_device_from_global_list
365 * remove the RCU, setup for a callback call,
366 * then wait for the callback to occur
368 static void del_edac_device_from_global_list(struct edac_device_ctl_info
371 list_del_rcu(&edac_device
->link
);
373 init_completion(&edac_device
->removal_complete
);
374 call_rcu(&edac_device
->rcu
, complete_edac_device_list_del
);
375 wait_for_completion(&edac_device
->removal_complete
);
379 * edac_device_workq_function
380 * performs the operation scheduled by a workq request
382 * this workq is embedded within an edac_device_ctl_info
383 * structure, that needs to be polled for possible error events.
385 * This operation is to acquire the list mutex lock
386 * (thus preventing insertation or deletion)
387 * and then call the device's poll function IFF this device is
388 * running polled and there is a poll function defined.
390 static void edac_device_workq_function(struct work_struct
*work_req
)
392 struct delayed_work
*d_work
= (struct delayed_work
*)work_req
;
393 struct edac_device_ctl_info
*edac_dev
= to_edac_device_ctl_work(d_work
);
395 mutex_lock(&device_ctls_mutex
);
397 /* If we are being removed, bail out immediately */
398 if (edac_dev
->op_state
== OP_OFFLINE
) {
399 mutex_unlock(&device_ctls_mutex
);
403 /* Only poll controllers that are running polled and have a check */
404 if ((edac_dev
->op_state
== OP_RUNNING_POLL
) &&
405 (edac_dev
->edac_check
!= NULL
)) {
406 edac_dev
->edac_check(edac_dev
);
409 mutex_unlock(&device_ctls_mutex
);
411 /* Reschedule the workq for the next time period to start again
412 * if the number of msec is for 1 sec, then adjust to the next
413 * whole one second to save timers fireing all over the period
414 * between integral seconds
416 if (edac_dev
->poll_msec
== 1000)
417 queue_delayed_work(edac_workqueue
, &edac_dev
->work
,
418 round_jiffies_relative(edac_dev
->delay
));
420 queue_delayed_work(edac_workqueue
, &edac_dev
->work
,
425 * edac_device_workq_setup
426 * initialize a workq item for this edac_device instance
427 * passing in the new delay period in msec
429 void edac_device_workq_setup(struct edac_device_ctl_info
*edac_dev
,
432 debugf0("%s()\n", __func__
);
434 /* take the arg 'msec' and set it into the control structure
435 * to used in the time period calculation
436 * then calc the number of jiffies that represents
438 edac_dev
->poll_msec
= msec
;
439 edac_dev
->delay
= msecs_to_jiffies(msec
);
441 INIT_DELAYED_WORK(&edac_dev
->work
, edac_device_workq_function
);
443 /* optimize here for the 1 second case, which will be normal value, to
444 * fire ON the 1 second time event. This helps reduce all sorts of
445 * timers firing on sub-second basis, while they are happy
446 * to fire together on the 1 second exactly
448 if (edac_dev
->poll_msec
== 1000)
449 queue_delayed_work(edac_workqueue
, &edac_dev
->work
,
450 round_jiffies_relative(edac_dev
->delay
));
452 queue_delayed_work(edac_workqueue
, &edac_dev
->work
,
457 * edac_device_workq_teardown
458 * stop the workq processing on this edac_dev
460 void edac_device_workq_teardown(struct edac_device_ctl_info
*edac_dev
)
464 status
= cancel_delayed_work(&edac_dev
->work
);
466 /* workq instance might be running, wait for it */
467 flush_workqueue(edac_workqueue
);
472 * edac_device_reset_delay_period
474 * need to stop any outstanding workq queued up at this time
475 * because we will be resetting the sleep time.
476 * Then restart the workq on the new delay
478 void edac_device_reset_delay_period(struct edac_device_ctl_info
*edac_dev
,
481 /* cancel the current workq request, without the mutex lock */
482 edac_device_workq_teardown(edac_dev
);
484 /* acquire the mutex before doing the workq setup */
485 mutex_lock(&device_ctls_mutex
);
487 /* restart the workq request, with new delay value */
488 edac_device_workq_setup(edac_dev
, value
);
490 mutex_unlock(&device_ctls_mutex
);
494 * edac_device_add_device: Insert the 'edac_dev' structure into the
495 * edac_device global list and create sysfs entries associated with
496 * edac_device structure.
497 * @edac_device: pointer to the edac_device structure to be added to the list
498 * 'edac_device' structure.
504 int edac_device_add_device(struct edac_device_ctl_info
*edac_dev
)
506 debugf0("%s()\n", __func__
);
508 #ifdef CONFIG_EDAC_DEBUG
509 if (edac_debug_level
>= 3)
510 edac_device_dump_device(edac_dev
);
512 mutex_lock(&device_ctls_mutex
);
514 if (add_edac_dev_to_global_list(edac_dev
))
517 /* set load time so that error rate can be tracked */
518 edac_dev
->start_time
= jiffies
;
520 /* create this instance's sysfs entries */
521 if (edac_device_create_sysfs(edac_dev
)) {
522 edac_device_printk(edac_dev
, KERN_WARNING
,
523 "failed to create sysfs device\n");
527 /* If there IS a check routine, then we are running POLLED */
528 if (edac_dev
->edac_check
!= NULL
) {
529 /* This instance is NOW RUNNING */
530 edac_dev
->op_state
= OP_RUNNING_POLL
;
533 * enable workq processing on this instance,
534 * default = 1000 msec
536 edac_device_workq_setup(edac_dev
, 1000);
538 edac_dev
->op_state
= OP_RUNNING_INTERRUPT
;
541 /* Report action taken */
542 edac_device_printk(edac_dev
, KERN_INFO
,
543 "Giving out device to module '%s' controller "
544 "'%s': DEV '%s' (%s)\n",
547 edac_dev_name(edac_dev
),
548 edac_op_state_to_string(edac_dev
->op_state
));
550 mutex_unlock(&device_ctls_mutex
);
554 /* Some error, so remove the entry from the lsit */
555 del_edac_device_from_global_list(edac_dev
);
558 mutex_unlock(&device_ctls_mutex
);
561 EXPORT_SYMBOL_GPL(edac_device_add_device
);
564 * edac_device_del_device:
565 * Remove sysfs entries for specified edac_device structure and
566 * then remove edac_device structure from global list
569 * Pointer to 'struct device' representing edac_device
570 * structure to remove.
573 * Pointer to removed edac_device structure,
574 * OR NULL if device not found.
576 struct edac_device_ctl_info
*edac_device_del_device(struct device
*dev
)
578 struct edac_device_ctl_info
*edac_dev
;
580 debugf0("%s()\n", __func__
);
582 mutex_lock(&device_ctls_mutex
);
584 /* Find the structure on the list, if not there, then leave */
585 edac_dev
= find_edac_device_by_dev(dev
);
586 if (edac_dev
== NULL
) {
587 mutex_unlock(&device_ctls_mutex
);
591 /* mark this instance as OFFLINE */
592 edac_dev
->op_state
= OP_OFFLINE
;
594 /* deregister from global list */
595 del_edac_device_from_global_list(edac_dev
);
597 mutex_unlock(&device_ctls_mutex
);
599 /* clear workq processing on this instance */
600 edac_device_workq_teardown(edac_dev
);
602 /* Tear down the sysfs entries for this instance */
603 edac_device_remove_sysfs(edac_dev
);
605 edac_printk(KERN_INFO
, EDAC_MC
,
606 "Removed device %d for %s %s: DEV %s\n",
608 edac_dev
->mod_name
, edac_dev
->ctl_name
, edac_dev_name(edac_dev
));
612 EXPORT_SYMBOL_GPL(edac_device_del_device
);
614 static inline int edac_device_get_log_ce(struct edac_device_ctl_info
*edac_dev
)
616 return edac_dev
->log_ce
;
619 static inline int edac_device_get_log_ue(struct edac_device_ctl_info
*edac_dev
)
621 return edac_dev
->log_ue
;
624 static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
627 return edac_dev
->panic_on_ue
;
631 * edac_device_handle_ce
632 * perform a common output and handling of an 'edac_dev' CE event
634 void edac_device_handle_ce(struct edac_device_ctl_info
*edac_dev
,
635 int inst_nr
, int block_nr
, const char *msg
)
637 struct edac_device_instance
*instance
;
638 struct edac_device_block
*block
= NULL
;
640 if ((inst_nr
>= edac_dev
->nr_instances
) || (inst_nr
< 0)) {
641 edac_device_printk(edac_dev
, KERN_ERR
,
642 "INTERNAL ERROR: 'instance' out of range "
643 "(%d >= %d)\n", inst_nr
,
644 edac_dev
->nr_instances
);
648 instance
= edac_dev
->instances
+ inst_nr
;
650 if ((block_nr
>= instance
->nr_blocks
) || (block_nr
< 0)) {
651 edac_device_printk(edac_dev
, KERN_ERR
,
652 "INTERNAL ERROR: instance %d 'block' "
653 "out of range (%d >= %d)\n",
655 instance
->nr_blocks
);
659 if (instance
->nr_blocks
> 0) {
660 block
= instance
->blocks
+ block_nr
;
661 block
->counters
.ce_count
++;
664 /* Propogate the count up the 'totals' tree */
665 instance
->counters
.ce_count
++;
666 edac_dev
->counters
.ce_count
++;
668 if (edac_device_get_log_ce(edac_dev
))
669 edac_device_printk(edac_dev
, KERN_WARNING
,
670 "CE: %s instance: %s block: %s '%s'\n",
671 edac_dev
->ctl_name
, instance
->name
,
672 block
? block
->name
: "N/A", msg
);
674 EXPORT_SYMBOL_GPL(edac_device_handle_ce
);
677 * edac_device_handle_ue
678 * perform a common output and handling of an 'edac_dev' UE event
680 void edac_device_handle_ue(struct edac_device_ctl_info
*edac_dev
,
681 int inst_nr
, int block_nr
, const char *msg
)
683 struct edac_device_instance
*instance
;
684 struct edac_device_block
*block
= NULL
;
686 if ((inst_nr
>= edac_dev
->nr_instances
) || (inst_nr
< 0)) {
687 edac_device_printk(edac_dev
, KERN_ERR
,
688 "INTERNAL ERROR: 'instance' out of range "
689 "(%d >= %d)\n", inst_nr
,
690 edac_dev
->nr_instances
);
694 instance
= edac_dev
->instances
+ inst_nr
;
696 if ((block_nr
>= instance
->nr_blocks
) || (block_nr
< 0)) {
697 edac_device_printk(edac_dev
, KERN_ERR
,
698 "INTERNAL ERROR: instance %d 'block' "
699 "out of range (%d >= %d)\n",
701 instance
->nr_blocks
);
705 if (instance
->nr_blocks
> 0) {
706 block
= instance
->blocks
+ block_nr
;
707 block
->counters
.ue_count
++;
710 /* Propogate the count up the 'totals' tree */
711 instance
->counters
.ue_count
++;
712 edac_dev
->counters
.ue_count
++;
714 if (edac_device_get_log_ue(edac_dev
))
715 edac_device_printk(edac_dev
, KERN_EMERG
,
716 "UE: %s instance: %s block: %s '%s'\n",
717 edac_dev
->ctl_name
, instance
->name
,
718 block
? block
->name
: "N/A", msg
);
720 if (edac_device_get_panic_on_ue(edac_dev
))
721 panic("EDAC %s: UE instance: %s block %s '%s'\n",
722 edac_dev
->ctl_name
, instance
->name
,
723 block
? block
->name
: "N/A", msg
);
725 EXPORT_SYMBOL_GPL(edac_device_handle_ue
);