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 debugf3("\tedac_dev = %p dev_idx=%d \n", edac_dev
, edac_dev
->dev_idx
);
44 debugf4("\tedac_dev->edac_check = %p\n", edac_dev
->edac_check
);
45 debugf3("\tdev = %p\n", edac_dev
->dev
);
46 debugf3("\tmod_name:ctl_name = %s:%s\n",
47 edac_dev
->mod_name
, edac_dev
->ctl_name
);
48 debugf3("\tpvt_info = %p\n\n", edac_dev
->pvt_info
);
50 #endif /* CONFIG_EDAC_DEBUG */
54 * edac_device_alloc_ctl_info()
55 * Allocate a new edac device control info structure
57 * The control structure is allocated in complete chunk
58 * from the OS. It is in turn sub allocated to the
59 * various objects that compose the struture
61 * The structure has a 'nr_instance' array within itself.
62 * Each instance represents a major component
63 * Example: L1 cache and L2 cache are 2 instance components
65 * Within each instance is an array of 'nr_blocks' blockoffsets
67 struct edac_device_ctl_info
*edac_device_alloc_ctl_info(
69 char *edac_device_name
, unsigned nr_instances
,
70 char *edac_block_name
, unsigned nr_blocks
,
71 unsigned offset_value
, /* zero, 1, or other based offset */
72 struct edac_dev_sysfs_block_attribute
*attrib_spec
, unsigned nr_attrib
,
75 struct edac_device_ctl_info
*dev_ctl
;
76 struct edac_device_instance
*dev_inst
, *inst
;
77 struct edac_device_block
*dev_blk
, *blk_p
, *blk
;
78 struct edac_dev_sysfs_block_attribute
*dev_attrib
, *attrib_p
, *attrib
;
81 unsigned instance
, block
, attr
;
85 debugf4("%s() instances=%d blocks=%d\n",
86 __func__
, 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.
95 dev_ctl
= (struct edac_device_ctl_info
*)NULL
;
97 /* Calc the 'end' offset past end of ONE ctl_info structure
98 * which will become the start of the 'instance' array
100 dev_inst
= edac_align_ptr(&dev_ctl
[1], sizeof(*dev_inst
));
102 /* Calc the 'end' offset past the instance array within the ctl_info
103 * which will become the start of the block array
105 dev_blk
= edac_align_ptr(&dev_inst
[nr_instances
], sizeof(*dev_blk
));
107 /* Calc the 'end' offset past the dev_blk array
108 * which will become the start of the attrib array, if any.
110 count
= nr_instances
* nr_blocks
;
111 dev_attrib
= edac_align_ptr(&dev_blk
[count
], sizeof(*dev_attrib
));
113 /* Check for case of when an attribute array is specified */
115 /* calc how many nr_attrib we need */
118 /* Calc the 'end' offset past the attributes array */
119 pvt
= edac_align_ptr(&dev_attrib
[count
], sz_private
);
121 /* no attribute array specificed */
122 pvt
= edac_align_ptr(dev_attrib
, sz_private
);
125 /* 'pvt' now points to where the private data area is.
126 * At this point 'pvt' (like dev_inst,dev_blk and dev_attrib)
127 * is baselined at ZERO
129 total_size
= ((unsigned long)pvt
) + sz_private
;
131 /* Allocate the amount of memory for the set of control structures */
132 dev_ctl
= kzalloc(total_size
, GFP_KERNEL
);
136 /* Adjust pointers so they point within the actual memory we
137 * just allocated rather than an imaginary chunk of memory
138 * located at address 0.
139 * 'dev_ctl' points to REAL memory, while the others are
140 * ZERO based and thus need to be adjusted to point within
141 * the allocated memory.
143 dev_inst
= (struct edac_device_instance
*)
144 (((char *)dev_ctl
) + ((unsigned long)dev_inst
));
145 dev_blk
= (struct edac_device_block
*)
146 (((char *)dev_ctl
) + ((unsigned long)dev_blk
));
147 dev_attrib
= (struct edac_dev_sysfs_block_attribute
*)
148 (((char *)dev_ctl
) + ((unsigned long)dev_attrib
));
149 pvt
= sz_private
? (((char *)dev_ctl
) + ((unsigned long)pvt
)) : NULL
;
151 /* Begin storing the information into the control info structure */
152 dev_ctl
->dev_idx
= device_index
;
153 dev_ctl
->nr_instances
= nr_instances
;
154 dev_ctl
->instances
= dev_inst
;
155 dev_ctl
->pvt_info
= pvt
;
157 /* Default logging of CEs and UEs */
161 /* Name of this edac device */
162 snprintf(dev_ctl
->name
,sizeof(dev_ctl
->name
),"%s",edac_device_name
);
164 debugf4("%s() edac_dev=%p next after end=%p\n",
165 __func__
, dev_ctl
, pvt
+ sz_private
);
167 /* Initialize every Instance */
168 for (instance
= 0; instance
< nr_instances
; instance
++) {
169 inst
= &dev_inst
[instance
];
171 inst
->nr_blocks
= nr_blocks
;
172 blk_p
= &dev_blk
[instance
* nr_blocks
];
173 inst
->blocks
= blk_p
;
175 /* name of this instance */
176 snprintf(inst
->name
, sizeof(inst
->name
),
177 "%s%u", edac_device_name
, instance
);
179 /* Initialize every block in each instance */
180 for (block
= 0; block
< nr_blocks
; block
++) {
182 blk
->instance
= inst
;
183 snprintf(blk
->name
, sizeof(blk
->name
),
184 "%s%d", edac_block_name
, block
+offset_value
);
186 debugf4("%s() instance=%d inst_p=%p block=#%d "
187 "block_p=%p name='%s'\n",
188 __func__
, instance
, inst
, block
,
191 /* if there are NO attributes OR no attribute pointer
192 * then continue on to next block iteration
194 if ((nr_attrib
== 0) || (attrib_spec
== NULL
))
197 /* setup the attribute array for this block */
198 blk
->nr_attribs
= nr_attrib
;
199 attrib_p
= &dev_attrib
[block
*nr_instances
*nr_attrib
];
200 blk
->block_attributes
= attrib_p
;
202 debugf4("%s() THIS BLOCK_ATTRIB=%p\n",
203 __func__
, blk
->block_attributes
);
205 /* Initialize every user specified attribute in this
206 * block with the data the caller passed in
207 * Each block gets its own copy of pointers,
208 * and its unique 'value'
210 for (attr
= 0; attr
< nr_attrib
; attr
++) {
211 attrib
= &attrib_p
[attr
];
213 /* populate the unique per attrib
214 * with the code pointers and info
216 attrib
->attr
= attrib_spec
[attr
].attr
;
217 attrib
->show
= attrib_spec
[attr
].show
;
218 attrib
->store
= attrib_spec
[attr
].store
;
220 attrib
->block
= blk
; /* up link */
222 debugf4("%s() alloc-attrib=%p attrib_name='%s' "
223 "attrib-spec=%p spec-name=%s\n",
224 __func__
, attrib
, attrib
->attr
.name
,
226 attrib_spec
[attr
].attr
.name
232 /* Mark this instance as merely ALLOCATED */
233 dev_ctl
->op_state
= OP_ALLOC
;
236 * Initialize the 'root' kobj for the edac_device controller
238 err
= edac_device_register_sysfs_main_kobj(dev_ctl
);
244 /* at this point, the root kobj is valid, and in order to
245 * 'free' the object, then the function:
246 * edac_device_unregister_sysfs_main_kobj() must be called
247 * which will perform kobj unregistration and the actual free
248 * will occur during the kobject callback operation
253 EXPORT_SYMBOL_GPL(edac_device_alloc_ctl_info
);
256 * edac_device_free_ctl_info()
257 * frees the memory allocated by the edac_device_alloc_ctl_info()
260 void edac_device_free_ctl_info(struct edac_device_ctl_info
*ctl_info
)
262 edac_device_unregister_sysfs_main_kobj(ctl_info
);
264 EXPORT_SYMBOL_GPL(edac_device_free_ctl_info
);
267 * find_edac_device_by_dev
268 * scans the edac_device list for a specific 'struct device *'
270 * lock to be held prior to call: device_ctls_mutex
273 * pointer to control structure managing 'dev'
274 * NULL if not found on list
276 static struct edac_device_ctl_info
*find_edac_device_by_dev(struct device
*dev
)
278 struct edac_device_ctl_info
*edac_dev
;
279 struct list_head
*item
;
281 debugf0("%s()\n", __func__
);
283 list_for_each(item
, &edac_device_list
) {
284 edac_dev
= list_entry(item
, struct edac_device_ctl_info
, link
);
286 if (edac_dev
->dev
== dev
)
294 * add_edac_dev_to_global_list
295 * Before calling this function, caller must
296 * assign a unique value to edac_dev->dev_idx.
298 * lock to be held prior to call: device_ctls_mutex
304 static int add_edac_dev_to_global_list(struct edac_device_ctl_info
*edac_dev
)
306 struct list_head
*item
, *insert_before
;
307 struct edac_device_ctl_info
*rover
;
309 insert_before
= &edac_device_list
;
311 /* Determine if already on the list */
312 rover
= find_edac_device_by_dev(edac_dev
->dev
);
313 if (unlikely(rover
!= NULL
))
316 /* Insert in ascending order by 'dev_idx', so find position */
317 list_for_each(item
, &edac_device_list
) {
318 rover
= list_entry(item
, struct edac_device_ctl_info
, link
);
320 if (rover
->dev_idx
>= edac_dev
->dev_idx
) {
321 if (unlikely(rover
->dev_idx
== edac_dev
->dev_idx
))
324 insert_before
= item
;
329 list_add_tail_rcu(&edac_dev
->link
, insert_before
);
333 edac_printk(KERN_WARNING
, EDAC_MC
,
334 "%s (%s) %s %s already assigned %d\n",
335 dev_name(rover
->dev
), edac_dev_name(rover
),
336 rover
->mod_name
, rover
->ctl_name
, rover
->dev_idx
);
340 edac_printk(KERN_WARNING
, EDAC_MC
,
341 "bug in low-level driver: attempt to assign\n"
342 " duplicate dev_idx %d in %s()\n", rover
->dev_idx
,
348 * del_edac_device_from_global_list
350 static void del_edac_device_from_global_list(struct edac_device_ctl_info
353 list_del_rcu(&edac_device
->link
);
355 /* these are for safe removal of devices from global list while
356 * NMI handlers may be traversing list
359 INIT_LIST_HEAD(&edac_device
->link
);
363 * edac_device_workq_function
364 * performs the operation scheduled by a workq request
366 * this workq is embedded within an edac_device_ctl_info
367 * structure, that needs to be polled for possible error events.
369 * This operation is to acquire the list mutex lock
370 * (thus preventing insertation or deletion)
371 * and then call the device's poll function IFF this device is
372 * running polled and there is a poll function defined.
374 static void edac_device_workq_function(struct work_struct
*work_req
)
376 struct delayed_work
*d_work
= to_delayed_work(work_req
);
377 struct edac_device_ctl_info
*edac_dev
= to_edac_device_ctl_work(d_work
);
379 mutex_lock(&device_ctls_mutex
);
381 /* If we are being removed, bail out immediately */
382 if (edac_dev
->op_state
== OP_OFFLINE
) {
383 mutex_unlock(&device_ctls_mutex
);
387 /* Only poll controllers that are running polled and have a check */
388 if ((edac_dev
->op_state
== OP_RUNNING_POLL
) &&
389 (edac_dev
->edac_check
!= NULL
)) {
390 edac_dev
->edac_check(edac_dev
);
393 mutex_unlock(&device_ctls_mutex
);
395 /* Reschedule the workq for the next time period to start again
396 * if the number of msec is for 1 sec, then adjust to the next
397 * whole one second to save timers fireing all over the period
398 * between integral seconds
400 if (edac_dev
->poll_msec
== 1000)
401 queue_delayed_work(edac_workqueue
, &edac_dev
->work
,
402 round_jiffies_relative(edac_dev
->delay
));
404 queue_delayed_work(edac_workqueue
, &edac_dev
->work
,
409 * edac_device_workq_setup
410 * initialize a workq item for this edac_device instance
411 * passing in the new delay period in msec
413 void edac_device_workq_setup(struct edac_device_ctl_info
*edac_dev
,
416 debugf0("%s()\n", __func__
);
418 /* take the arg 'msec' and set it into the control structure
419 * to used in the time period calculation
420 * then calc the number of jiffies that represents
422 edac_dev
->poll_msec
= msec
;
423 edac_dev
->delay
= msecs_to_jiffies(msec
);
425 INIT_DELAYED_WORK(&edac_dev
->work
, edac_device_workq_function
);
427 /* optimize here for the 1 second case, which will be normal value, to
428 * fire ON the 1 second time event. This helps reduce all sorts of
429 * timers firing on sub-second basis, while they are happy
430 * to fire together on the 1 second exactly
432 if (edac_dev
->poll_msec
== 1000)
433 queue_delayed_work(edac_workqueue
, &edac_dev
->work
,
434 round_jiffies_relative(edac_dev
->delay
));
436 queue_delayed_work(edac_workqueue
, &edac_dev
->work
,
441 * edac_device_workq_teardown
442 * stop the workq processing on this edac_dev
444 void edac_device_workq_teardown(struct edac_device_ctl_info
*edac_dev
)
448 status
= cancel_delayed_work(&edac_dev
->work
);
450 /* workq instance might be running, wait for it */
451 flush_workqueue(edac_workqueue
);
456 * edac_device_reset_delay_period
458 * need to stop any outstanding workq queued up at this time
459 * because we will be resetting the sleep time.
460 * Then restart the workq on the new delay
462 void edac_device_reset_delay_period(struct edac_device_ctl_info
*edac_dev
,
465 /* cancel the current workq request, without the mutex lock */
466 edac_device_workq_teardown(edac_dev
);
468 /* acquire the mutex before doing the workq setup */
469 mutex_lock(&device_ctls_mutex
);
471 /* restart the workq request, with new delay value */
472 edac_device_workq_setup(edac_dev
, value
);
474 mutex_unlock(&device_ctls_mutex
);
478 * edac_device_alloc_index: Allocate a unique device index number
481 * allocated index number
483 int edac_device_alloc_index(void)
485 static atomic_t device_indexes
= ATOMIC_INIT(0);
487 return atomic_inc_return(&device_indexes
) - 1;
489 EXPORT_SYMBOL_GPL(edac_device_alloc_index
);
492 * edac_device_add_device: Insert the 'edac_dev' structure into the
493 * edac_device global list and create sysfs entries associated with
494 * edac_device structure.
495 * @edac_device: pointer to the edac_device structure to be added to the list
496 * 'edac_device' structure.
502 int edac_device_add_device(struct edac_device_ctl_info
*edac_dev
)
504 debugf0("%s()\n", __func__
);
506 #ifdef CONFIG_EDAC_DEBUG
507 if (edac_debug_level
>= 3)
508 edac_device_dump_device(edac_dev
);
510 mutex_lock(&device_ctls_mutex
);
512 if (add_edac_dev_to_global_list(edac_dev
))
515 /* set load time so that error rate can be tracked */
516 edac_dev
->start_time
= jiffies
;
518 /* create this instance's sysfs entries */
519 if (edac_device_create_sysfs(edac_dev
)) {
520 edac_device_printk(edac_dev
, KERN_WARNING
,
521 "failed to create sysfs device\n");
525 /* If there IS a check routine, then we are running POLLED */
526 if (edac_dev
->edac_check
!= NULL
) {
527 /* This instance is NOW RUNNING */
528 edac_dev
->op_state
= OP_RUNNING_POLL
;
531 * enable workq processing on this instance,
532 * default = 1000 msec
534 edac_device_workq_setup(edac_dev
, 1000);
536 edac_dev
->op_state
= OP_RUNNING_INTERRUPT
;
539 /* Report action taken */
540 edac_device_printk(edac_dev
, KERN_INFO
,
541 "Giving out device to module '%s' controller "
542 "'%s': DEV '%s' (%s)\n",
545 edac_dev_name(edac_dev
),
546 edac_op_state_to_string(edac_dev
->op_state
));
548 mutex_unlock(&device_ctls_mutex
);
552 /* Some error, so remove the entry from the lsit */
553 del_edac_device_from_global_list(edac_dev
);
556 mutex_unlock(&device_ctls_mutex
);
559 EXPORT_SYMBOL_GPL(edac_device_add_device
);
562 * edac_device_del_device:
563 * Remove sysfs entries for specified edac_device structure and
564 * then remove edac_device structure from global list
567 * Pointer to 'struct device' representing edac_device
568 * structure to remove.
571 * Pointer to removed edac_device structure,
572 * OR NULL if device not found.
574 struct edac_device_ctl_info
*edac_device_del_device(struct device
*dev
)
576 struct edac_device_ctl_info
*edac_dev
;
578 debugf0("%s()\n", __func__
);
580 mutex_lock(&device_ctls_mutex
);
582 /* Find the structure on the list, if not there, then leave */
583 edac_dev
= find_edac_device_by_dev(dev
);
584 if (edac_dev
== NULL
) {
585 mutex_unlock(&device_ctls_mutex
);
589 /* mark this instance as OFFLINE */
590 edac_dev
->op_state
= OP_OFFLINE
;
592 /* deregister from global list */
593 del_edac_device_from_global_list(edac_dev
);
595 mutex_unlock(&device_ctls_mutex
);
597 /* clear workq processing on this instance */
598 edac_device_workq_teardown(edac_dev
);
600 /* Tear down the sysfs entries for this instance */
601 edac_device_remove_sysfs(edac_dev
);
603 edac_printk(KERN_INFO
, EDAC_MC
,
604 "Removed device %d for %s %s: DEV %s\n",
606 edac_dev
->mod_name
, edac_dev
->ctl_name
, edac_dev_name(edac_dev
));
610 EXPORT_SYMBOL_GPL(edac_device_del_device
);
612 static inline int edac_device_get_log_ce(struct edac_device_ctl_info
*edac_dev
)
614 return edac_dev
->log_ce
;
617 static inline int edac_device_get_log_ue(struct edac_device_ctl_info
*edac_dev
)
619 return edac_dev
->log_ue
;
622 static inline int edac_device_get_panic_on_ue(struct edac_device_ctl_info
625 return edac_dev
->panic_on_ue
;
629 * edac_device_handle_ce
630 * perform a common output and handling of an 'edac_dev' CE event
632 void edac_device_handle_ce(struct edac_device_ctl_info
*edac_dev
,
633 int inst_nr
, int block_nr
, const char *msg
)
635 struct edac_device_instance
*instance
;
636 struct edac_device_block
*block
= NULL
;
638 if ((inst_nr
>= edac_dev
->nr_instances
) || (inst_nr
< 0)) {
639 edac_device_printk(edac_dev
, KERN_ERR
,
640 "INTERNAL ERROR: 'instance' out of range "
641 "(%d >= %d)\n", inst_nr
,
642 edac_dev
->nr_instances
);
646 instance
= edac_dev
->instances
+ inst_nr
;
648 if ((block_nr
>= instance
->nr_blocks
) || (block_nr
< 0)) {
649 edac_device_printk(edac_dev
, KERN_ERR
,
650 "INTERNAL ERROR: instance %d 'block' "
651 "out of range (%d >= %d)\n",
653 instance
->nr_blocks
);
657 if (instance
->nr_blocks
> 0) {
658 block
= instance
->blocks
+ block_nr
;
659 block
->counters
.ce_count
++;
662 /* Propagate the count up the 'totals' tree */
663 instance
->counters
.ce_count
++;
664 edac_dev
->counters
.ce_count
++;
666 if (edac_device_get_log_ce(edac_dev
))
667 edac_device_printk(edac_dev
, KERN_WARNING
,
668 "CE: %s instance: %s block: %s '%s'\n",
669 edac_dev
->ctl_name
, instance
->name
,
670 block
? block
->name
: "N/A", msg
);
672 EXPORT_SYMBOL_GPL(edac_device_handle_ce
);
675 * edac_device_handle_ue
676 * perform a common output and handling of an 'edac_dev' UE event
678 void edac_device_handle_ue(struct edac_device_ctl_info
*edac_dev
,
679 int inst_nr
, int block_nr
, const char *msg
)
681 struct edac_device_instance
*instance
;
682 struct edac_device_block
*block
= NULL
;
684 if ((inst_nr
>= edac_dev
->nr_instances
) || (inst_nr
< 0)) {
685 edac_device_printk(edac_dev
, KERN_ERR
,
686 "INTERNAL ERROR: 'instance' out of range "
687 "(%d >= %d)\n", inst_nr
,
688 edac_dev
->nr_instances
);
692 instance
= edac_dev
->instances
+ inst_nr
;
694 if ((block_nr
>= instance
->nr_blocks
) || (block_nr
< 0)) {
695 edac_device_printk(edac_dev
, KERN_ERR
,
696 "INTERNAL ERROR: instance %d 'block' "
697 "out of range (%d >= %d)\n",
699 instance
->nr_blocks
);
703 if (instance
->nr_blocks
> 0) {
704 block
= instance
->blocks
+ block_nr
;
705 block
->counters
.ue_count
++;
708 /* Propagate the count up the 'totals' tree */
709 instance
->counters
.ue_count
++;
710 edac_dev
->counters
.ue_count
++;
712 if (edac_device_get_log_ue(edac_dev
))
713 edac_device_printk(edac_dev
, KERN_EMERG
,
714 "UE: %s instance: %s block: %s '%s'\n",
715 edac_dev
->ctl_name
, instance
->name
,
716 block
? block
->name
: "N/A", msg
);
718 if (edac_device_get_panic_on_ue(edac_dev
))
719 panic("EDAC %s: UE instance: %s block %s '%s'\n",
720 edac_dev
->ctl_name
, instance
->name
,
721 block
? block
->name
: "N/A", msg
);
723 EXPORT_SYMBOL_GPL(edac_device_handle_ue
);