Pull ia64-clocksource into release branch
[GitHub/MotorolaMobilityLLC/kernel-slsi.git] / drivers / edac / edac_mc.c
1 /*
2 * edac_mc kernel module
3 * (C) 2005, 2006 Linux Networx (http://lnxi.com)
4 * This file may be distributed under the terms of the
5 * GNU General Public License.
6 *
7 * Written by Thayne Harbaugh
8 * Based on work by Dan Hollis <goemon at anime dot net> and others.
9 * http://www.anime.net/~goemon/linux-ecc/
10 *
11 * Modified by Dave Peterson and Doug Thompson
12 *
13 */
14
15 #include <linux/module.h>
16 #include <linux/proc_fs.h>
17 #include <linux/kernel.h>
18 #include <linux/types.h>
19 #include <linux/smp.h>
20 #include <linux/init.h>
21 #include <linux/sysctl.h>
22 #include <linux/highmem.h>
23 #include <linux/timer.h>
24 #include <linux/slab.h>
25 #include <linux/jiffies.h>
26 #include <linux/spinlock.h>
27 #include <linux/list.h>
28 #include <linux/sysdev.h>
29 #include <linux/ctype.h>
30 #include <linux/edac.h>
31 #include <asm/uaccess.h>
32 #include <asm/page.h>
33 #include <asm/edac.h>
34 #include "edac_core.h"
35 #include "edac_module.h"
36
37 /* lock to memory controller's control array */
38 static DEFINE_MUTEX(mem_ctls_mutex);
39 static struct list_head mc_devices = LIST_HEAD_INIT(mc_devices);
40
41 #ifdef CONFIG_EDAC_DEBUG
42
43 static void edac_mc_dump_channel(struct channel_info *chan)
44 {
45 debugf4("\tchannel = %p\n", chan);
46 debugf4("\tchannel->chan_idx = %d\n", chan->chan_idx);
47 debugf4("\tchannel->ce_count = %d\n", chan->ce_count);
48 debugf4("\tchannel->label = '%s'\n", chan->label);
49 debugf4("\tchannel->csrow = %p\n\n", chan->csrow);
50 }
51
52 static void edac_mc_dump_csrow(struct csrow_info *csrow)
53 {
54 debugf4("\tcsrow = %p\n", csrow);
55 debugf4("\tcsrow->csrow_idx = %d\n", csrow->csrow_idx);
56 debugf4("\tcsrow->first_page = 0x%lx\n", csrow->first_page);
57 debugf4("\tcsrow->last_page = 0x%lx\n", csrow->last_page);
58 debugf4("\tcsrow->page_mask = 0x%lx\n", csrow->page_mask);
59 debugf4("\tcsrow->nr_pages = 0x%x\n", csrow->nr_pages);
60 debugf4("\tcsrow->nr_channels = %d\n", csrow->nr_channels);
61 debugf4("\tcsrow->channels = %p\n", csrow->channels);
62 debugf4("\tcsrow->mci = %p\n\n", csrow->mci);
63 }
64
65 static void edac_mc_dump_mci(struct mem_ctl_info *mci)
66 {
67 debugf3("\tmci = %p\n", mci);
68 debugf3("\tmci->mtype_cap = %lx\n", mci->mtype_cap);
69 debugf3("\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap);
70 debugf3("\tmci->edac_cap = %lx\n", mci->edac_cap);
71 debugf4("\tmci->edac_check = %p\n", mci->edac_check);
72 debugf3("\tmci->nr_csrows = %d, csrows = %p\n",
73 mci->nr_csrows, mci->csrows);
74 debugf3("\tdev = %p\n", mci->dev);
75 debugf3("\tmod_name:ctl_name = %s:%s\n", mci->mod_name, mci->ctl_name);
76 debugf3("\tpvt_info = %p\n\n", mci->pvt_info);
77 }
78
79 #endif /* CONFIG_EDAC_DEBUG */
80
81 /* 'ptr' points to a possibly unaligned item X such that sizeof(X) is 'size'.
82 * Adjust 'ptr' so that its alignment is at least as stringent as what the
83 * compiler would provide for X and return the aligned result.
84 *
85 * If 'size' is a constant, the compiler will optimize this whole function
86 * down to either a no-op or the addition of a constant to the value of 'ptr'.
87 */
88 void *edac_align_ptr(void *ptr, unsigned size)
89 {
90 unsigned align, r;
91
92 /* Here we assume that the alignment of a "long long" is the most
93 * stringent alignment that the compiler will ever provide by default.
94 * As far as I know, this is a reasonable assumption.
95 */
96 if (size > sizeof(long))
97 align = sizeof(long long);
98 else if (size > sizeof(int))
99 align = sizeof(long);
100 else if (size > sizeof(short))
101 align = sizeof(int);
102 else if (size > sizeof(char))
103 align = sizeof(short);
104 else
105 return (char *)ptr;
106
107 r = size % align;
108
109 if (r == 0)
110 return (char *)ptr;
111
112 return (void *)(((unsigned long)ptr) + align - r);
113 }
114
115 /**
116 * edac_mc_alloc: Allocate a struct mem_ctl_info structure
117 * @size_pvt: size of private storage needed
118 * @nr_csrows: Number of CWROWS needed for this MC
119 * @nr_chans: Number of channels for the MC
120 *
121 * Everything is kmalloc'ed as one big chunk - more efficient.
122 * Only can be used if all structures have the same lifetime - otherwise
123 * you have to allocate and initialize your own structures.
124 *
125 * Use edac_mc_free() to free mc structures allocated by this function.
126 *
127 * Returns:
128 * NULL allocation failed
129 * struct mem_ctl_info pointer
130 */
131 struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows,
132 unsigned nr_chans, int edac_index)
133 {
134 struct mem_ctl_info *mci;
135 struct csrow_info *csi, *csrow;
136 struct channel_info *chi, *chp, *chan;
137 void *pvt;
138 unsigned size;
139 int row, chn;
140 int err;
141
142 /* Figure out the offsets of the various items from the start of an mc
143 * structure. We want the alignment of each item to be at least as
144 * stringent as what the compiler would provide if we could simply
145 * hardcode everything into a single struct.
146 */
147 mci = (struct mem_ctl_info *)0;
148 csi = edac_align_ptr(&mci[1], sizeof(*csi));
149 chi = edac_align_ptr(&csi[nr_csrows], sizeof(*chi));
150 pvt = edac_align_ptr(&chi[nr_chans * nr_csrows], sz_pvt);
151 size = ((unsigned long)pvt) + sz_pvt;
152
153 mci = kzalloc(size, GFP_KERNEL);
154 if (mci == NULL)
155 return NULL;
156
157 /* Adjust pointers so they point within the memory we just allocated
158 * rather than an imaginary chunk of memory located at address 0.
159 */
160 csi = (struct csrow_info *)(((char *)mci) + ((unsigned long)csi));
161 chi = (struct channel_info *)(((char *)mci) + ((unsigned long)chi));
162 pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL;
163
164 /* setup index and various internal pointers */
165 mci->mc_idx = edac_index;
166 mci->csrows = csi;
167 mci->pvt_info = pvt;
168 mci->nr_csrows = nr_csrows;
169
170 for (row = 0; row < nr_csrows; row++) {
171 csrow = &csi[row];
172 csrow->csrow_idx = row;
173 csrow->mci = mci;
174 csrow->nr_channels = nr_chans;
175 chp = &chi[row * nr_chans];
176 csrow->channels = chp;
177
178 for (chn = 0; chn < nr_chans; chn++) {
179 chan = &chp[chn];
180 chan->chan_idx = chn;
181 chan->csrow = csrow;
182 }
183 }
184
185 mci->op_state = OP_ALLOC;
186
187 /*
188 * Initialize the 'root' kobj for the edac_mc controller
189 */
190 err = edac_mc_register_sysfs_main_kobj(mci);
191 if (err) {
192 kfree(mci);
193 return NULL;
194 }
195
196 /* at this point, the root kobj is valid, and in order to
197 * 'free' the object, then the function:
198 * edac_mc_unregister_sysfs_main_kobj() must be called
199 * which will perform kobj unregistration and the actual free
200 * will occur during the kobject callback operation
201 */
202 return mci;
203 }
204 EXPORT_SYMBOL_GPL(edac_mc_alloc);
205
206 /**
207 * edac_mc_free
208 * 'Free' a previously allocated 'mci' structure
209 * @mci: pointer to a struct mem_ctl_info structure
210 */
211 void edac_mc_free(struct mem_ctl_info *mci)
212 {
213 edac_mc_unregister_sysfs_main_kobj(mci);
214 }
215 EXPORT_SYMBOL_GPL(edac_mc_free);
216
217 static struct mem_ctl_info *find_mci_by_dev(struct device *dev)
218 {
219 struct mem_ctl_info *mci;
220 struct list_head *item;
221
222 debugf3("%s()\n", __func__);
223
224 list_for_each(item, &mc_devices) {
225 mci = list_entry(item, struct mem_ctl_info, link);
226
227 if (mci->dev == dev)
228 return mci;
229 }
230
231 return NULL;
232 }
233
234 /*
235 * handler for EDAC to check if NMI type handler has asserted interrupt
236 */
237 static int edac_mc_assert_error_check_and_clear(void)
238 {
239 int old_state;
240
241 if (edac_op_state == EDAC_OPSTATE_POLL)
242 return 1;
243
244 old_state = edac_err_assert;
245 edac_err_assert = 0;
246
247 return old_state;
248 }
249
250 /*
251 * edac_mc_workq_function
252 * performs the operation scheduled by a workq request
253 */
254 static void edac_mc_workq_function(struct work_struct *work_req)
255 {
256 struct delayed_work *d_work = (struct delayed_work *)work_req;
257 struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
258
259 mutex_lock(&mem_ctls_mutex);
260
261 /* if this control struct has movd to offline state, we are done */
262 if (mci->op_state == OP_OFFLINE) {
263 mutex_unlock(&mem_ctls_mutex);
264 return;
265 }
266
267 /* Only poll controllers that are running polled and have a check */
268 if (edac_mc_assert_error_check_and_clear() && (mci->edac_check != NULL))
269 mci->edac_check(mci);
270
271 /*
272 * FIXME: temp place holder for PCI checks,
273 * goes away when we break out PCI
274 */
275 edac_pci_do_parity_check();
276
277 mutex_unlock(&mem_ctls_mutex);
278
279 /* Reschedule */
280 queue_delayed_work(edac_workqueue, &mci->work,
281 msecs_to_jiffies(edac_mc_get_poll_msec()));
282 }
283
284 /*
285 * edac_mc_workq_setup
286 * initialize a workq item for this mci
287 * passing in the new delay period in msec
288 *
289 * locking model:
290 *
291 * called with the mem_ctls_mutex held
292 */
293 static void edac_mc_workq_setup(struct mem_ctl_info *mci, unsigned msec)
294 {
295 debugf0("%s()\n", __func__);
296
297 /* if this instance is not in the POLL state, then simply return */
298 if (mci->op_state != OP_RUNNING_POLL)
299 return;
300
301 INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
302 queue_delayed_work(edac_workqueue, &mci->work, msecs_to_jiffies(msec));
303 }
304
305 /*
306 * edac_mc_workq_teardown
307 * stop the workq processing on this mci
308 *
309 * locking model:
310 *
311 * called WITHOUT lock held
312 */
313 static void edac_mc_workq_teardown(struct mem_ctl_info *mci)
314 {
315 int status;
316
317 /* if not running POLL, leave now */
318 if (mci->op_state == OP_RUNNING_POLL) {
319 status = cancel_delayed_work(&mci->work);
320 if (status == 0) {
321 debugf0("%s() not canceled, flush the queue\n",
322 __func__);
323
324 /* workq instance might be running, wait for it */
325 flush_workqueue(edac_workqueue);
326 }
327 }
328 }
329
330 /*
331 * edac_reset_delay_period
332 */
333 static void edac_reset_delay_period(struct mem_ctl_info *mci, unsigned long value)
334 {
335 /* cancel the current workq request */
336 edac_mc_workq_teardown(mci);
337
338 /* lock the list of devices for the new setup */
339 mutex_lock(&mem_ctls_mutex);
340
341 /* restart the workq request, with new delay value */
342 edac_mc_workq_setup(mci, value);
343
344 mutex_unlock(&mem_ctls_mutex);
345 }
346
347 /* Return 0 on success, 1 on failure.
348 * Before calling this function, caller must
349 * assign a unique value to mci->mc_idx.
350 *
351 * locking model:
352 *
353 * called with the mem_ctls_mutex lock held
354 */
355 static int add_mc_to_global_list(struct mem_ctl_info *mci)
356 {
357 struct list_head *item, *insert_before;
358 struct mem_ctl_info *p;
359
360 insert_before = &mc_devices;
361
362 p = find_mci_by_dev(mci->dev);
363 if (unlikely(p != NULL))
364 goto fail0;
365
366 list_for_each(item, &mc_devices) {
367 p = list_entry(item, struct mem_ctl_info, link);
368
369 if (p->mc_idx >= mci->mc_idx) {
370 if (unlikely(p->mc_idx == mci->mc_idx))
371 goto fail1;
372
373 insert_before = item;
374 break;
375 }
376 }
377
378 list_add_tail_rcu(&mci->link, insert_before);
379 atomic_inc(&edac_handlers);
380 return 0;
381
382 fail0:
383 edac_printk(KERN_WARNING, EDAC_MC,
384 "%s (%s) %s %s already assigned %d\n", p->dev->bus_id,
385 dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
386 return 1;
387
388 fail1:
389 edac_printk(KERN_WARNING, EDAC_MC,
390 "bug in low-level driver: attempt to assign\n"
391 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
392 return 1;
393 }
394
395 static void complete_mc_list_del(struct rcu_head *head)
396 {
397 struct mem_ctl_info *mci;
398
399 mci = container_of(head, struct mem_ctl_info, rcu);
400 INIT_LIST_HEAD(&mci->link);
401 complete(&mci->complete);
402 }
403
404 static void del_mc_from_global_list(struct mem_ctl_info *mci)
405 {
406 atomic_dec(&edac_handlers);
407 list_del_rcu(&mci->link);
408 init_completion(&mci->complete);
409 call_rcu(&mci->rcu, complete_mc_list_del);
410 wait_for_completion(&mci->complete);
411 }
412
413 /**
414 * edac_mc_find: Search for a mem_ctl_info structure whose index is 'idx'.
415 *
416 * If found, return a pointer to the structure.
417 * Else return NULL.
418 *
419 * Caller must hold mem_ctls_mutex.
420 */
421 struct mem_ctl_info *edac_mc_find(int idx)
422 {
423 struct list_head *item;
424 struct mem_ctl_info *mci;
425
426 list_for_each(item, &mc_devices) {
427 mci = list_entry(item, struct mem_ctl_info, link);
428
429 if (mci->mc_idx >= idx) {
430 if (mci->mc_idx == idx)
431 return mci;
432
433 break;
434 }
435 }
436
437 return NULL;
438 }
439 EXPORT_SYMBOL(edac_mc_find);
440
441 /**
442 * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and
443 * create sysfs entries associated with mci structure
444 * @mci: pointer to the mci structure to be added to the list
445 * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure.
446 *
447 * Return:
448 * 0 Success
449 * !0 Failure
450 */
451
452 /* FIXME - should a warning be printed if no error detection? correction? */
453 int edac_mc_add_mc(struct mem_ctl_info *mci)
454 {
455 debugf0("%s()\n", __func__);
456
457 #ifdef CONFIG_EDAC_DEBUG
458 if (edac_debug_level >= 3)
459 edac_mc_dump_mci(mci);
460
461 if (edac_debug_level >= 4) {
462 int i;
463
464 for (i = 0; i < mci->nr_csrows; i++) {
465 int j;
466
467 edac_mc_dump_csrow(&mci->csrows[i]);
468 for (j = 0; j < mci->csrows[i].nr_channels; j++)
469 edac_mc_dump_channel(&mci->csrows[i].
470 channels[j]);
471 }
472 }
473 #endif
474 mutex_lock(&mem_ctls_mutex);
475
476 if (add_mc_to_global_list(mci))
477 goto fail0;
478
479 /* set load time so that error rate can be tracked */
480 mci->start_time = jiffies;
481
482 if (edac_create_sysfs_mci_device(mci)) {
483 edac_mc_printk(mci, KERN_WARNING,
484 "failed to create sysfs device\n");
485 goto fail1;
486 }
487
488 /* If there IS a check routine, then we are running POLLED */
489 if (mci->edac_check != NULL) {
490 /* This instance is NOW RUNNING */
491 mci->op_state = OP_RUNNING_POLL;
492
493 edac_mc_workq_setup(mci, edac_mc_get_poll_msec());
494 } else {
495 mci->op_state = OP_RUNNING_INTERRUPT;
496 }
497
498 /* Report action taken */
499 edac_mc_printk(mci, KERN_INFO, "Giving out device to '%s' '%s':"
500 " DEV %s\n", mci->mod_name, mci->ctl_name, dev_name(mci));
501
502 mutex_unlock(&mem_ctls_mutex);
503 return 0;
504
505 fail1:
506 del_mc_from_global_list(mci);
507
508 fail0:
509 mutex_unlock(&mem_ctls_mutex);
510 return 1;
511 }
512 EXPORT_SYMBOL_GPL(edac_mc_add_mc);
513
514 /**
515 * edac_mc_del_mc: Remove sysfs entries for specified mci structure and
516 * remove mci structure from global list
517 * @pdev: Pointer to 'struct device' representing mci structure to remove.
518 *
519 * Return pointer to removed mci structure, or NULL if device not found.
520 */
521 struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
522 {
523 struct mem_ctl_info *mci;
524
525 debugf0("%s()\n", __func__);
526
527 mutex_lock(&mem_ctls_mutex);
528
529 /* find the requested mci struct in the global list */
530 mci = find_mci_by_dev(dev);
531 if (mci == NULL) {
532 mutex_unlock(&mem_ctls_mutex);
533 return NULL;
534 }
535
536 /* marking MCI offline */
537 mci->op_state = OP_OFFLINE;
538
539 del_mc_from_global_list(mci);
540 mutex_unlock(&mem_ctls_mutex);
541
542 /* flush workq processes and remove sysfs */
543 edac_mc_workq_teardown(mci);
544 edac_remove_sysfs_mci_device(mci);
545
546 edac_printk(KERN_INFO, EDAC_MC,
547 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
548 mci->mod_name, mci->ctl_name, dev_name(mci));
549
550 return mci;
551 }
552 EXPORT_SYMBOL_GPL(edac_mc_del_mc);
553
554 static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
555 u32 size)
556 {
557 struct page *pg;
558 void *virt_addr;
559 unsigned long flags = 0;
560
561 debugf3("%s()\n", __func__);
562
563 /* ECC error page was not in our memory. Ignore it. */
564 if (!pfn_valid(page))
565 return;
566
567 /* Find the actual page structure then map it and fix */
568 pg = pfn_to_page(page);
569
570 if (PageHighMem(pg))
571 local_irq_save(flags);
572
573 virt_addr = kmap_atomic(pg, KM_BOUNCE_READ);
574
575 /* Perform architecture specific atomic scrub operation */
576 atomic_scrub(virt_addr + offset, size);
577
578 /* Unmap and complete */
579 kunmap_atomic(virt_addr, KM_BOUNCE_READ);
580
581 if (PageHighMem(pg))
582 local_irq_restore(flags);
583 }
584
585 /* FIXME - should return -1 */
586 int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
587 {
588 struct csrow_info *csrows = mci->csrows;
589 int row, i;
590
591 debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page);
592 row = -1;
593
594 for (i = 0; i < mci->nr_csrows; i++) {
595 struct csrow_info *csrow = &csrows[i];
596
597 if (csrow->nr_pages == 0)
598 continue;
599
600 debugf3("MC%d: %s(): first(0x%lx) page(0x%lx) last(0x%lx) "
601 "mask(0x%lx)\n", mci->mc_idx, __func__,
602 csrow->first_page, page, csrow->last_page,
603 csrow->page_mask);
604
605 if ((page >= csrow->first_page) &&
606 (page <= csrow->last_page) &&
607 ((page & csrow->page_mask) ==
608 (csrow->first_page & csrow->page_mask))) {
609 row = i;
610 break;
611 }
612 }
613
614 if (row == -1)
615 edac_mc_printk(mci, KERN_ERR,
616 "could not look up page error address %lx\n",
617 (unsigned long)page);
618
619 return row;
620 }
621 EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
622
623 /* FIXME - setable log (warning/emerg) levels */
624 /* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */
625 void edac_mc_handle_ce(struct mem_ctl_info *mci,
626 unsigned long page_frame_number,
627 unsigned long offset_in_page, unsigned long syndrome,
628 int row, int channel, const char *msg)
629 {
630 unsigned long remapped_page;
631
632 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
633
634 /* FIXME - maybe make panic on INTERNAL ERROR an option */
635 if (row >= mci->nr_csrows || row < 0) {
636 /* something is wrong */
637 edac_mc_printk(mci, KERN_ERR,
638 "INTERNAL ERROR: row out of range "
639 "(%d >= %d)\n", row, mci->nr_csrows);
640 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
641 return;
642 }
643
644 if (channel >= mci->csrows[row].nr_channels || channel < 0) {
645 /* something is wrong */
646 edac_mc_printk(mci, KERN_ERR,
647 "INTERNAL ERROR: channel out of range "
648 "(%d >= %d)\n", channel,
649 mci->csrows[row].nr_channels);
650 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
651 return;
652 }
653
654 if (edac_mc_get_log_ce())
655 /* FIXME - put in DIMM location */
656 edac_mc_printk(mci, KERN_WARNING,
657 "CE page 0x%lx, offset 0x%lx, grain %d, syndrome "
658 "0x%lx, row %d, channel %d, label \"%s\": %s\n",
659 page_frame_number, offset_in_page,
660 mci->csrows[row].grain, syndrome, row, channel,
661 mci->csrows[row].channels[channel].label, msg);
662
663 mci->ce_count++;
664 mci->csrows[row].ce_count++;
665 mci->csrows[row].channels[channel].ce_count++;
666
667 if (mci->scrub_mode & SCRUB_SW_SRC) {
668 /*
669 * Some MC's can remap memory so that it is still available
670 * at a different address when PCI devices map into memory.
671 * MC's that can't do this lose the memory where PCI devices
672 * are mapped. This mapping is MC dependant and so we call
673 * back into the MC driver for it to map the MC page to
674 * a physical (CPU) page which can then be mapped to a virtual
675 * page - which can then be scrubbed.
676 */
677 remapped_page = mci->ctl_page_to_phys ?
678 mci->ctl_page_to_phys(mci, page_frame_number) :
679 page_frame_number;
680
681 edac_mc_scrub_block(remapped_page, offset_in_page,
682 mci->csrows[row].grain);
683 }
684 }
685 EXPORT_SYMBOL_GPL(edac_mc_handle_ce);
686
687 void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg)
688 {
689 if (edac_mc_get_log_ce())
690 edac_mc_printk(mci, KERN_WARNING,
691 "CE - no information available: %s\n", msg);
692
693 mci->ce_noinfo_count++;
694 mci->ce_count++;
695 }
696 EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info);
697
698 void edac_mc_handle_ue(struct mem_ctl_info *mci,
699 unsigned long page_frame_number,
700 unsigned long offset_in_page, int row, const char *msg)
701 {
702 int len = EDAC_MC_LABEL_LEN * 4;
703 char labels[len + 1];
704 char *pos = labels;
705 int chan;
706 int chars;
707
708 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
709
710 /* FIXME - maybe make panic on INTERNAL ERROR an option */
711 if (row >= mci->nr_csrows || row < 0) {
712 /* something is wrong */
713 edac_mc_printk(mci, KERN_ERR,
714 "INTERNAL ERROR: row out of range "
715 "(%d >= %d)\n", row, mci->nr_csrows);
716 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
717 return;
718 }
719
720 chars = snprintf(pos, len + 1, "%s",
721 mci->csrows[row].channels[0].label);
722 len -= chars;
723 pos += chars;
724
725 for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0);
726 chan++) {
727 chars = snprintf(pos, len + 1, ":%s",
728 mci->csrows[row].channels[chan].label);
729 len -= chars;
730 pos += chars;
731 }
732
733 if (edac_mc_get_log_ue())
734 edac_mc_printk(mci, KERN_EMERG,
735 "UE page 0x%lx, offset 0x%lx, grain %d, row %d, "
736 "labels \"%s\": %s\n", page_frame_number,
737 offset_in_page, mci->csrows[row].grain, row,
738 labels, msg);
739
740 if (edac_mc_get_panic_on_ue())
741 panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, "
742 "row %d, labels \"%s\": %s\n", mci->mc_idx,
743 page_frame_number, offset_in_page,
744 mci->csrows[row].grain, row, labels, msg);
745
746 mci->ue_count++;
747 mci->csrows[row].ue_count++;
748 }
749 EXPORT_SYMBOL_GPL(edac_mc_handle_ue);
750
751 void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg)
752 {
753 if (edac_mc_get_panic_on_ue())
754 panic("EDAC MC%d: Uncorrected Error", mci->mc_idx);
755
756 if (edac_mc_get_log_ue())
757 edac_mc_printk(mci, KERN_WARNING,
758 "UE - no information available: %s\n", msg);
759 mci->ue_noinfo_count++;
760 mci->ue_count++;
761 }
762 EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info);
763
764 /*************************************************************
765 * On Fully Buffered DIMM modules, this help function is
766 * called to process UE events
767 */
768 void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci,
769 unsigned int csrow,
770 unsigned int channela,
771 unsigned int channelb, char *msg)
772 {
773 int len = EDAC_MC_LABEL_LEN * 4;
774 char labels[len + 1];
775 char *pos = labels;
776 int chars;
777
778 if (csrow >= mci->nr_csrows) {
779 /* something is wrong */
780 edac_mc_printk(mci, KERN_ERR,
781 "INTERNAL ERROR: row out of range (%d >= %d)\n",
782 csrow, mci->nr_csrows);
783 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
784 return;
785 }
786
787 if (channela >= mci->csrows[csrow].nr_channels) {
788 /* something is wrong */
789 edac_mc_printk(mci, KERN_ERR,
790 "INTERNAL ERROR: channel-a out of range "
791 "(%d >= %d)\n",
792 channela, mci->csrows[csrow].nr_channels);
793 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
794 return;
795 }
796
797 if (channelb >= mci->csrows[csrow].nr_channels) {
798 /* something is wrong */
799 edac_mc_printk(mci, KERN_ERR,
800 "INTERNAL ERROR: channel-b out of range "
801 "(%d >= %d)\n",
802 channelb, mci->csrows[csrow].nr_channels);
803 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
804 return;
805 }
806
807 mci->ue_count++;
808 mci->csrows[csrow].ue_count++;
809
810 /* Generate the DIMM labels from the specified channels */
811 chars = snprintf(pos, len + 1, "%s",
812 mci->csrows[csrow].channels[channela].label);
813 len -= chars;
814 pos += chars;
815 chars = snprintf(pos, len + 1, "-%s",
816 mci->csrows[csrow].channels[channelb].label);
817
818 if (edac_mc_get_log_ue())
819 edac_mc_printk(mci, KERN_EMERG,
820 "UE row %d, channel-a= %d channel-b= %d "
821 "labels \"%s\": %s\n", csrow, channela, channelb,
822 labels, msg);
823
824 if (edac_mc_get_panic_on_ue())
825 panic("UE row %d, channel-a= %d channel-b= %d "
826 "labels \"%s\": %s\n", csrow, channela,
827 channelb, labels, msg);
828 }
829 EXPORT_SYMBOL(edac_mc_handle_fbd_ue);
830
831 /*************************************************************
832 * On Fully Buffered DIMM modules, this help function is
833 * called to process CE events
834 */
835 void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci,
836 unsigned int csrow, unsigned int channel, char *msg)
837 {
838
839 /* Ensure boundary values */
840 if (csrow >= mci->nr_csrows) {
841 /* something is wrong */
842 edac_mc_printk(mci, KERN_ERR,
843 "INTERNAL ERROR: row out of range (%d >= %d)\n",
844 csrow, mci->nr_csrows);
845 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
846 return;
847 }
848 if (channel >= mci->csrows[csrow].nr_channels) {
849 /* something is wrong */
850 edac_mc_printk(mci, KERN_ERR,
851 "INTERNAL ERROR: channel out of range (%d >= %d)\n",
852 channel, mci->csrows[csrow].nr_channels);
853 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
854 return;
855 }
856
857 if (edac_mc_get_log_ce())
858 /* FIXME - put in DIMM location */
859 edac_mc_printk(mci, KERN_WARNING,
860 "CE row %d, channel %d, label \"%s\": %s\n",
861 csrow, channel,
862 mci->csrows[csrow].channels[channel].label, msg);
863
864 mci->ce_count++;
865 mci->csrows[csrow].ce_count++;
866 mci->csrows[csrow].channels[channel].ce_count++;
867 }
868 EXPORT_SYMBOL(edac_mc_handle_fbd_ce);
869
870 /*
871 * Iterate over all MC instances and check for ECC, et al, errors
872 */
873 void edac_check_mc_devices(void)
874 {
875 struct list_head *item;
876 struct mem_ctl_info *mci;
877
878 debugf3("%s()\n", __func__);
879 mutex_lock(&mem_ctls_mutex);
880
881 list_for_each(item, &mc_devices) {
882 mci = list_entry(item, struct mem_ctl_info, link);
883
884 if (mci->edac_check != NULL)
885 mci->edac_check(mci);
886 }
887
888 mutex_unlock(&mem_ctls_mutex);
889 }