Merge branch 'drm-intel-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/ickle...
[deliverable/linux.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 LIST_HEAD(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 /*
80 * keep those in sync with the enum mem_type
81 */
82 const char *edac_mem_types[] = {
83 "Empty csrow",
84 "Reserved csrow type",
85 "Unknown csrow type",
86 "Fast page mode RAM",
87 "Extended data out RAM",
88 "Burst Extended data out RAM",
89 "Single data rate SDRAM",
90 "Registered single data rate SDRAM",
91 "Double data rate SDRAM",
92 "Registered Double data rate SDRAM",
93 "Rambus DRAM",
94 "Unbuffered DDR2 RAM",
95 "Fully buffered DDR2",
96 "Registered DDR2 RAM",
97 "Rambus XDR",
98 "Unbuffered DDR3 RAM",
99 "Registered DDR3 RAM",
100 };
101 EXPORT_SYMBOL_GPL(edac_mem_types);
102
103 #endif /* CONFIG_EDAC_DEBUG */
104
105 /* 'ptr' points to a possibly unaligned item X such that sizeof(X) is 'size'.
106 * Adjust 'ptr' so that its alignment is at least as stringent as what the
107 * compiler would provide for X and return the aligned result.
108 *
109 * If 'size' is a constant, the compiler will optimize this whole function
110 * down to either a no-op or the addition of a constant to the value of 'ptr'.
111 */
112 void *edac_align_ptr(void *ptr, unsigned size)
113 {
114 unsigned align, r;
115
116 /* Here we assume that the alignment of a "long long" is the most
117 * stringent alignment that the compiler will ever provide by default.
118 * As far as I know, this is a reasonable assumption.
119 */
120 if (size > sizeof(long))
121 align = sizeof(long long);
122 else if (size > sizeof(int))
123 align = sizeof(long);
124 else if (size > sizeof(short))
125 align = sizeof(int);
126 else if (size > sizeof(char))
127 align = sizeof(short);
128 else
129 return (char *)ptr;
130
131 r = size % align;
132
133 if (r == 0)
134 return (char *)ptr;
135
136 return (void *)(((unsigned long)ptr) + align - r);
137 }
138
139 /**
140 * edac_mc_alloc: Allocate a struct mem_ctl_info structure
141 * @size_pvt: size of private storage needed
142 * @nr_csrows: Number of CWROWS needed for this MC
143 * @nr_chans: Number of channels for the MC
144 *
145 * Everything is kmalloc'ed as one big chunk - more efficient.
146 * Only can be used if all structures have the same lifetime - otherwise
147 * you have to allocate and initialize your own structures.
148 *
149 * Use edac_mc_free() to free mc structures allocated by this function.
150 *
151 * Returns:
152 * NULL allocation failed
153 * struct mem_ctl_info pointer
154 */
155 struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows,
156 unsigned nr_chans, int edac_index)
157 {
158 struct mem_ctl_info *mci;
159 struct csrow_info *csi, *csrow;
160 struct channel_info *chi, *chp, *chan;
161 void *pvt;
162 unsigned size;
163 int row, chn;
164 int err;
165
166 /* Figure out the offsets of the various items from the start of an mc
167 * structure. We want the alignment of each item to be at least as
168 * stringent as what the compiler would provide if we could simply
169 * hardcode everything into a single struct.
170 */
171 mci = (struct mem_ctl_info *)0;
172 csi = edac_align_ptr(&mci[1], sizeof(*csi));
173 chi = edac_align_ptr(&csi[nr_csrows], sizeof(*chi));
174 pvt = edac_align_ptr(&chi[nr_chans * nr_csrows], sz_pvt);
175 size = ((unsigned long)pvt) + sz_pvt;
176
177 mci = kzalloc(size, GFP_KERNEL);
178 if (mci == NULL)
179 return NULL;
180
181 /* Adjust pointers so they point within the memory we just allocated
182 * rather than an imaginary chunk of memory located at address 0.
183 */
184 csi = (struct csrow_info *)(((char *)mci) + ((unsigned long)csi));
185 chi = (struct channel_info *)(((char *)mci) + ((unsigned long)chi));
186 pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL;
187
188 /* setup index and various internal pointers */
189 mci->mc_idx = edac_index;
190 mci->csrows = csi;
191 mci->pvt_info = pvt;
192 mci->nr_csrows = nr_csrows;
193
194 for (row = 0; row < nr_csrows; row++) {
195 csrow = &csi[row];
196 csrow->csrow_idx = row;
197 csrow->mci = mci;
198 csrow->nr_channels = nr_chans;
199 chp = &chi[row * nr_chans];
200 csrow->channels = chp;
201
202 for (chn = 0; chn < nr_chans; chn++) {
203 chan = &chp[chn];
204 chan->chan_idx = chn;
205 chan->csrow = csrow;
206 }
207 }
208
209 mci->op_state = OP_ALLOC;
210
211 /*
212 * Initialize the 'root' kobj for the edac_mc controller
213 */
214 err = edac_mc_register_sysfs_main_kobj(mci);
215 if (err) {
216 kfree(mci);
217 return NULL;
218 }
219
220 /* at this point, the root kobj is valid, and in order to
221 * 'free' the object, then the function:
222 * edac_mc_unregister_sysfs_main_kobj() must be called
223 * which will perform kobj unregistration and the actual free
224 * will occur during the kobject callback operation
225 */
226 return mci;
227 }
228 EXPORT_SYMBOL_GPL(edac_mc_alloc);
229
230 /**
231 * edac_mc_free
232 * 'Free' a previously allocated 'mci' structure
233 * @mci: pointer to a struct mem_ctl_info structure
234 */
235 void edac_mc_free(struct mem_ctl_info *mci)
236 {
237 edac_mc_unregister_sysfs_main_kobj(mci);
238 }
239 EXPORT_SYMBOL_GPL(edac_mc_free);
240
241
242 /*
243 * find_mci_by_dev
244 *
245 * scan list of controllers looking for the one that manages
246 * the 'dev' device
247 */
248 static struct mem_ctl_info *find_mci_by_dev(struct device *dev)
249 {
250 struct mem_ctl_info *mci;
251 struct list_head *item;
252
253 debugf3("%s()\n", __func__);
254
255 list_for_each(item, &mc_devices) {
256 mci = list_entry(item, struct mem_ctl_info, link);
257
258 if (mci->dev == dev)
259 return mci;
260 }
261
262 return NULL;
263 }
264
265 /*
266 * handler for EDAC to check if NMI type handler has asserted interrupt
267 */
268 static int edac_mc_assert_error_check_and_clear(void)
269 {
270 int old_state;
271
272 if (edac_op_state == EDAC_OPSTATE_POLL)
273 return 1;
274
275 old_state = edac_err_assert;
276 edac_err_assert = 0;
277
278 return old_state;
279 }
280
281 /*
282 * edac_mc_workq_function
283 * performs the operation scheduled by a workq request
284 */
285 static void edac_mc_workq_function(struct work_struct *work_req)
286 {
287 struct delayed_work *d_work = to_delayed_work(work_req);
288 struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
289
290 mutex_lock(&mem_ctls_mutex);
291
292 /* if this control struct has movd to offline state, we are done */
293 if (mci->op_state == OP_OFFLINE) {
294 mutex_unlock(&mem_ctls_mutex);
295 return;
296 }
297
298 /* Only poll controllers that are running polled and have a check */
299 if (edac_mc_assert_error_check_and_clear() && (mci->edac_check != NULL))
300 mci->edac_check(mci);
301
302 mutex_unlock(&mem_ctls_mutex);
303
304 /* Reschedule */
305 queue_delayed_work(edac_workqueue, &mci->work,
306 msecs_to_jiffies(edac_mc_get_poll_msec()));
307 }
308
309 /*
310 * edac_mc_workq_setup
311 * initialize a workq item for this mci
312 * passing in the new delay period in msec
313 *
314 * locking model:
315 *
316 * called with the mem_ctls_mutex held
317 */
318 static void edac_mc_workq_setup(struct mem_ctl_info *mci, unsigned msec)
319 {
320 debugf0("%s()\n", __func__);
321
322 /* if this instance is not in the POLL state, then simply return */
323 if (mci->op_state != OP_RUNNING_POLL)
324 return;
325
326 INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
327 queue_delayed_work(edac_workqueue, &mci->work, msecs_to_jiffies(msec));
328 }
329
330 /*
331 * edac_mc_workq_teardown
332 * stop the workq processing on this mci
333 *
334 * locking model:
335 *
336 * called WITHOUT lock held
337 */
338 static void edac_mc_workq_teardown(struct mem_ctl_info *mci)
339 {
340 int status;
341
342 if (mci->op_state != OP_RUNNING_POLL)
343 return;
344
345 status = cancel_delayed_work(&mci->work);
346 if (status == 0) {
347 debugf0("%s() not canceled, flush the queue\n",
348 __func__);
349
350 /* workq instance might be running, wait for it */
351 flush_workqueue(edac_workqueue);
352 }
353 }
354
355 /*
356 * edac_mc_reset_delay_period(unsigned long value)
357 *
358 * user space has updated our poll period value, need to
359 * reset our workq delays
360 */
361 void edac_mc_reset_delay_period(int value)
362 {
363 struct mem_ctl_info *mci;
364 struct list_head *item;
365
366 mutex_lock(&mem_ctls_mutex);
367
368 /* scan the list and turn off all workq timers, doing so under lock
369 */
370 list_for_each(item, &mc_devices) {
371 mci = list_entry(item, struct mem_ctl_info, link);
372
373 if (mci->op_state == OP_RUNNING_POLL)
374 cancel_delayed_work(&mci->work);
375 }
376
377 mutex_unlock(&mem_ctls_mutex);
378
379
380 /* re-walk the list, and reset the poll delay */
381 mutex_lock(&mem_ctls_mutex);
382
383 list_for_each(item, &mc_devices) {
384 mci = list_entry(item, struct mem_ctl_info, link);
385
386 edac_mc_workq_setup(mci, (unsigned long) value);
387 }
388
389 mutex_unlock(&mem_ctls_mutex);
390 }
391
392
393
394 /* Return 0 on success, 1 on failure.
395 * Before calling this function, caller must
396 * assign a unique value to mci->mc_idx.
397 *
398 * locking model:
399 *
400 * called with the mem_ctls_mutex lock held
401 */
402 static int add_mc_to_global_list(struct mem_ctl_info *mci)
403 {
404 struct list_head *item, *insert_before;
405 struct mem_ctl_info *p;
406
407 insert_before = &mc_devices;
408
409 p = find_mci_by_dev(mci->dev);
410 if (unlikely(p != NULL))
411 goto fail0;
412
413 list_for_each(item, &mc_devices) {
414 p = list_entry(item, struct mem_ctl_info, link);
415
416 if (p->mc_idx >= mci->mc_idx) {
417 if (unlikely(p->mc_idx == mci->mc_idx))
418 goto fail1;
419
420 insert_before = item;
421 break;
422 }
423 }
424
425 list_add_tail_rcu(&mci->link, insert_before);
426 atomic_inc(&edac_handlers);
427 return 0;
428
429 fail0:
430 edac_printk(KERN_WARNING, EDAC_MC,
431 "%s (%s) %s %s already assigned %d\n", dev_name(p->dev),
432 edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
433 return 1;
434
435 fail1:
436 edac_printk(KERN_WARNING, EDAC_MC,
437 "bug in low-level driver: attempt to assign\n"
438 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
439 return 1;
440 }
441
442 static void complete_mc_list_del(struct rcu_head *head)
443 {
444 struct mem_ctl_info *mci;
445
446 mci = container_of(head, struct mem_ctl_info, rcu);
447 INIT_LIST_HEAD(&mci->link);
448 }
449
450 static void del_mc_from_global_list(struct mem_ctl_info *mci)
451 {
452 atomic_dec(&edac_handlers);
453 list_del_rcu(&mci->link);
454 call_rcu(&mci->rcu, complete_mc_list_del);
455 rcu_barrier();
456 }
457
458 /**
459 * edac_mc_find: Search for a mem_ctl_info structure whose index is 'idx'.
460 *
461 * If found, return a pointer to the structure.
462 * Else return NULL.
463 *
464 * Caller must hold mem_ctls_mutex.
465 */
466 struct mem_ctl_info *edac_mc_find(int idx)
467 {
468 struct list_head *item;
469 struct mem_ctl_info *mci;
470
471 list_for_each(item, &mc_devices) {
472 mci = list_entry(item, struct mem_ctl_info, link);
473
474 if (mci->mc_idx >= idx) {
475 if (mci->mc_idx == idx)
476 return mci;
477
478 break;
479 }
480 }
481
482 return NULL;
483 }
484 EXPORT_SYMBOL(edac_mc_find);
485
486 /**
487 * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and
488 * create sysfs entries associated with mci structure
489 * @mci: pointer to the mci structure to be added to the list
490 * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure.
491 *
492 * Return:
493 * 0 Success
494 * !0 Failure
495 */
496
497 /* FIXME - should a warning be printed if no error detection? correction? */
498 int edac_mc_add_mc(struct mem_ctl_info *mci)
499 {
500 debugf0("%s()\n", __func__);
501
502 #ifdef CONFIG_EDAC_DEBUG
503 if (edac_debug_level >= 3)
504 edac_mc_dump_mci(mci);
505
506 if (edac_debug_level >= 4) {
507 int i;
508
509 for (i = 0; i < mci->nr_csrows; i++) {
510 int j;
511
512 edac_mc_dump_csrow(&mci->csrows[i]);
513 for (j = 0; j < mci->csrows[i].nr_channels; j++)
514 edac_mc_dump_channel(&mci->csrows[i].
515 channels[j]);
516 }
517 }
518 #endif
519 mutex_lock(&mem_ctls_mutex);
520
521 if (add_mc_to_global_list(mci))
522 goto fail0;
523
524 /* set load time so that error rate can be tracked */
525 mci->start_time = jiffies;
526
527 if (edac_create_sysfs_mci_device(mci)) {
528 edac_mc_printk(mci, KERN_WARNING,
529 "failed to create sysfs device\n");
530 goto fail1;
531 }
532
533 /* If there IS a check routine, then we are running POLLED */
534 if (mci->edac_check != NULL) {
535 /* This instance is NOW RUNNING */
536 mci->op_state = OP_RUNNING_POLL;
537
538 edac_mc_workq_setup(mci, edac_mc_get_poll_msec());
539 } else {
540 mci->op_state = OP_RUNNING_INTERRUPT;
541 }
542
543 /* Report action taken */
544 edac_mc_printk(mci, KERN_INFO, "Giving out device to '%s' '%s':"
545 " DEV %s\n", mci->mod_name, mci->ctl_name, edac_dev_name(mci));
546
547 mutex_unlock(&mem_ctls_mutex);
548 return 0;
549
550 fail1:
551 del_mc_from_global_list(mci);
552
553 fail0:
554 mutex_unlock(&mem_ctls_mutex);
555 return 1;
556 }
557 EXPORT_SYMBOL_GPL(edac_mc_add_mc);
558
559 /**
560 * edac_mc_del_mc: Remove sysfs entries for specified mci structure and
561 * remove mci structure from global list
562 * @pdev: Pointer to 'struct device' representing mci structure to remove.
563 *
564 * Return pointer to removed mci structure, or NULL if device not found.
565 */
566 struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
567 {
568 struct mem_ctl_info *mci;
569
570 debugf0("%s()\n", __func__);
571
572 mutex_lock(&mem_ctls_mutex);
573
574 /* find the requested mci struct in the global list */
575 mci = find_mci_by_dev(dev);
576 if (mci == NULL) {
577 mutex_unlock(&mem_ctls_mutex);
578 return NULL;
579 }
580
581 /* marking MCI offline */
582 mci->op_state = OP_OFFLINE;
583
584 del_mc_from_global_list(mci);
585 mutex_unlock(&mem_ctls_mutex);
586
587 /* flush workq processes and remove sysfs */
588 edac_mc_workq_teardown(mci);
589 edac_remove_sysfs_mci_device(mci);
590
591 edac_printk(KERN_INFO, EDAC_MC,
592 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
593 mci->mod_name, mci->ctl_name, edac_dev_name(mci));
594
595 return mci;
596 }
597 EXPORT_SYMBOL_GPL(edac_mc_del_mc);
598
599 static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
600 u32 size)
601 {
602 struct page *pg;
603 void *virt_addr;
604 unsigned long flags = 0;
605
606 debugf3("%s()\n", __func__);
607
608 /* ECC error page was not in our memory. Ignore it. */
609 if (!pfn_valid(page))
610 return;
611
612 /* Find the actual page structure then map it and fix */
613 pg = pfn_to_page(page);
614
615 if (PageHighMem(pg))
616 local_irq_save(flags);
617
618 virt_addr = kmap_atomic(pg, KM_BOUNCE_READ);
619
620 /* Perform architecture specific atomic scrub operation */
621 atomic_scrub(virt_addr + offset, size);
622
623 /* Unmap and complete */
624 kunmap_atomic(virt_addr, KM_BOUNCE_READ);
625
626 if (PageHighMem(pg))
627 local_irq_restore(flags);
628 }
629
630 /* FIXME - should return -1 */
631 int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
632 {
633 struct csrow_info *csrows = mci->csrows;
634 int row, i;
635
636 debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page);
637 row = -1;
638
639 for (i = 0; i < mci->nr_csrows; i++) {
640 struct csrow_info *csrow = &csrows[i];
641
642 if (csrow->nr_pages == 0)
643 continue;
644
645 debugf3("MC%d: %s(): first(0x%lx) page(0x%lx) last(0x%lx) "
646 "mask(0x%lx)\n", mci->mc_idx, __func__,
647 csrow->first_page, page, csrow->last_page,
648 csrow->page_mask);
649
650 if ((page >= csrow->first_page) &&
651 (page <= csrow->last_page) &&
652 ((page & csrow->page_mask) ==
653 (csrow->first_page & csrow->page_mask))) {
654 row = i;
655 break;
656 }
657 }
658
659 if (row == -1)
660 edac_mc_printk(mci, KERN_ERR,
661 "could not look up page error address %lx\n",
662 (unsigned long)page);
663
664 return row;
665 }
666 EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
667
668 /* FIXME - setable log (warning/emerg) levels */
669 /* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */
670 void edac_mc_handle_ce(struct mem_ctl_info *mci,
671 unsigned long page_frame_number,
672 unsigned long offset_in_page, unsigned long syndrome,
673 int row, int channel, const char *msg)
674 {
675 unsigned long remapped_page;
676
677 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
678
679 /* FIXME - maybe make panic on INTERNAL ERROR an option */
680 if (row >= mci->nr_csrows || row < 0) {
681 /* something is wrong */
682 edac_mc_printk(mci, KERN_ERR,
683 "INTERNAL ERROR: row out of range "
684 "(%d >= %d)\n", row, mci->nr_csrows);
685 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
686 return;
687 }
688
689 if (channel >= mci->csrows[row].nr_channels || channel < 0) {
690 /* something is wrong */
691 edac_mc_printk(mci, KERN_ERR,
692 "INTERNAL ERROR: channel out of range "
693 "(%d >= %d)\n", channel,
694 mci->csrows[row].nr_channels);
695 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
696 return;
697 }
698
699 if (edac_mc_get_log_ce())
700 /* FIXME - put in DIMM location */
701 edac_mc_printk(mci, KERN_WARNING,
702 "CE page 0x%lx, offset 0x%lx, grain %d, syndrome "
703 "0x%lx, row %d, channel %d, label \"%s\": %s\n",
704 page_frame_number, offset_in_page,
705 mci->csrows[row].grain, syndrome, row, channel,
706 mci->csrows[row].channels[channel].label, msg);
707
708 mci->ce_count++;
709 mci->csrows[row].ce_count++;
710 mci->csrows[row].channels[channel].ce_count++;
711
712 if (mci->scrub_mode & SCRUB_SW_SRC) {
713 /*
714 * Some MC's can remap memory so that it is still available
715 * at a different address when PCI devices map into memory.
716 * MC's that can't do this lose the memory where PCI devices
717 * are mapped. This mapping is MC dependant and so we call
718 * back into the MC driver for it to map the MC page to
719 * a physical (CPU) page which can then be mapped to a virtual
720 * page - which can then be scrubbed.
721 */
722 remapped_page = mci->ctl_page_to_phys ?
723 mci->ctl_page_to_phys(mci, page_frame_number) :
724 page_frame_number;
725
726 edac_mc_scrub_block(remapped_page, offset_in_page,
727 mci->csrows[row].grain);
728 }
729 }
730 EXPORT_SYMBOL_GPL(edac_mc_handle_ce);
731
732 void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg)
733 {
734 if (edac_mc_get_log_ce())
735 edac_mc_printk(mci, KERN_WARNING,
736 "CE - no information available: %s\n", msg);
737
738 mci->ce_noinfo_count++;
739 mci->ce_count++;
740 }
741 EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info);
742
743 void edac_mc_handle_ue(struct mem_ctl_info *mci,
744 unsigned long page_frame_number,
745 unsigned long offset_in_page, int row, const char *msg)
746 {
747 int len = EDAC_MC_LABEL_LEN * 4;
748 char labels[len + 1];
749 char *pos = labels;
750 int chan;
751 int chars;
752
753 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
754
755 /* FIXME - maybe make panic on INTERNAL ERROR an option */
756 if (row >= mci->nr_csrows || row < 0) {
757 /* something is wrong */
758 edac_mc_printk(mci, KERN_ERR,
759 "INTERNAL ERROR: row out of range "
760 "(%d >= %d)\n", row, mci->nr_csrows);
761 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
762 return;
763 }
764
765 chars = snprintf(pos, len + 1, "%s",
766 mci->csrows[row].channels[0].label);
767 len -= chars;
768 pos += chars;
769
770 for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0);
771 chan++) {
772 chars = snprintf(pos, len + 1, ":%s",
773 mci->csrows[row].channels[chan].label);
774 len -= chars;
775 pos += chars;
776 }
777
778 if (edac_mc_get_log_ue())
779 edac_mc_printk(mci, KERN_EMERG,
780 "UE page 0x%lx, offset 0x%lx, grain %d, row %d, "
781 "labels \"%s\": %s\n", page_frame_number,
782 offset_in_page, mci->csrows[row].grain, row,
783 labels, msg);
784
785 if (edac_mc_get_panic_on_ue())
786 panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, "
787 "row %d, labels \"%s\": %s\n", mci->mc_idx,
788 page_frame_number, offset_in_page,
789 mci->csrows[row].grain, row, labels, msg);
790
791 mci->ue_count++;
792 mci->csrows[row].ue_count++;
793 }
794 EXPORT_SYMBOL_GPL(edac_mc_handle_ue);
795
796 void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg)
797 {
798 if (edac_mc_get_panic_on_ue())
799 panic("EDAC MC%d: Uncorrected Error", mci->mc_idx);
800
801 if (edac_mc_get_log_ue())
802 edac_mc_printk(mci, KERN_WARNING,
803 "UE - no information available: %s\n", msg);
804 mci->ue_noinfo_count++;
805 mci->ue_count++;
806 }
807 EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info);
808
809 /*************************************************************
810 * On Fully Buffered DIMM modules, this help function is
811 * called to process UE events
812 */
813 void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci,
814 unsigned int csrow,
815 unsigned int channela,
816 unsigned int channelb, char *msg)
817 {
818 int len = EDAC_MC_LABEL_LEN * 4;
819 char labels[len + 1];
820 char *pos = labels;
821 int chars;
822
823 if (csrow >= mci->nr_csrows) {
824 /* something is wrong */
825 edac_mc_printk(mci, KERN_ERR,
826 "INTERNAL ERROR: row out of range (%d >= %d)\n",
827 csrow, mci->nr_csrows);
828 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
829 return;
830 }
831
832 if (channela >= mci->csrows[csrow].nr_channels) {
833 /* something is wrong */
834 edac_mc_printk(mci, KERN_ERR,
835 "INTERNAL ERROR: channel-a out of range "
836 "(%d >= %d)\n",
837 channela, mci->csrows[csrow].nr_channels);
838 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
839 return;
840 }
841
842 if (channelb >= mci->csrows[csrow].nr_channels) {
843 /* something is wrong */
844 edac_mc_printk(mci, KERN_ERR,
845 "INTERNAL ERROR: channel-b out of range "
846 "(%d >= %d)\n",
847 channelb, mci->csrows[csrow].nr_channels);
848 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
849 return;
850 }
851
852 mci->ue_count++;
853 mci->csrows[csrow].ue_count++;
854
855 /* Generate the DIMM labels from the specified channels */
856 chars = snprintf(pos, len + 1, "%s",
857 mci->csrows[csrow].channels[channela].label);
858 len -= chars;
859 pos += chars;
860 chars = snprintf(pos, len + 1, "-%s",
861 mci->csrows[csrow].channels[channelb].label);
862
863 if (edac_mc_get_log_ue())
864 edac_mc_printk(mci, KERN_EMERG,
865 "UE row %d, channel-a= %d channel-b= %d "
866 "labels \"%s\": %s\n", csrow, channela, channelb,
867 labels, msg);
868
869 if (edac_mc_get_panic_on_ue())
870 panic("UE row %d, channel-a= %d channel-b= %d "
871 "labels \"%s\": %s\n", csrow, channela,
872 channelb, labels, msg);
873 }
874 EXPORT_SYMBOL(edac_mc_handle_fbd_ue);
875
876 /*************************************************************
877 * On Fully Buffered DIMM modules, this help function is
878 * called to process CE events
879 */
880 void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci,
881 unsigned int csrow, unsigned int channel, char *msg)
882 {
883
884 /* Ensure boundary values */
885 if (csrow >= mci->nr_csrows) {
886 /* something is wrong */
887 edac_mc_printk(mci, KERN_ERR,
888 "INTERNAL ERROR: row out of range (%d >= %d)\n",
889 csrow, mci->nr_csrows);
890 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
891 return;
892 }
893 if (channel >= mci->csrows[csrow].nr_channels) {
894 /* something is wrong */
895 edac_mc_printk(mci, KERN_ERR,
896 "INTERNAL ERROR: channel out of range (%d >= %d)\n",
897 channel, mci->csrows[csrow].nr_channels);
898 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
899 return;
900 }
901
902 if (edac_mc_get_log_ce())
903 /* FIXME - put in DIMM location */
904 edac_mc_printk(mci, KERN_WARNING,
905 "CE row %d, channel %d, label \"%s\": %s\n",
906 csrow, channel,
907 mci->csrows[csrow].channels[channel].label, msg);
908
909 mci->ce_count++;
910 mci->csrows[csrow].ce_count++;
911 mci->csrows[csrow].channels[channel].ce_count++;
912 }
913 EXPORT_SYMBOL(edac_mc_handle_fbd_ce);
This page took 0.072546 seconds and 5 git commands to generate.