]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/edac/edac_mc.c
be2net: Bug fix in init code in probe
[net-next-2.6.git] / drivers / edac / edac_mc.c
CommitLineData
da9bb1d2
AC
1/*
2 * edac_mc kernel module
49c0dab7 3 * (C) 2005, 2006 Linux Networx (http://lnxi.com)
da9bb1d2
AC
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
da9bb1d2
AC
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>
c0d12172 30#include <linux/edac.h>
da9bb1d2
AC
31#include <asm/uaccess.h>
32#include <asm/page.h>
33#include <asm/edac.h>
20bcb7a8 34#include "edac_core.h"
7c9281d7 35#include "edac_module.h"
da9bb1d2 36
da9bb1d2 37/* lock to memory controller's control array */
63b7df91 38static DEFINE_MUTEX(mem_ctls_mutex);
ff6ac2a6 39static LIST_HEAD(mc_devices);
da9bb1d2 40
da9bb1d2
AC
41#ifdef CONFIG_EDAC_DEBUG
42
2da1c119 43static void edac_mc_dump_channel(struct channel_info *chan)
da9bb1d2
AC
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
2da1c119 52static void edac_mc_dump_csrow(struct csrow_info *csrow)
da9bb1d2
AC
53{
54 debugf4("\tcsrow = %p\n", csrow);
55 debugf4("\tcsrow->csrow_idx = %d\n", csrow->csrow_idx);
079708b9 56 debugf4("\tcsrow->first_page = 0x%lx\n", csrow->first_page);
da9bb1d2
AC
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);
079708b9 60 debugf4("\tcsrow->nr_channels = %d\n", csrow->nr_channels);
da9bb1d2
AC
61 debugf4("\tcsrow->channels = %p\n", csrow->channels);
62 debugf4("\tcsrow->mci = %p\n\n", csrow->mci);
63}
64
2da1c119 65static void edac_mc_dump_mci(struct mem_ctl_info *mci)
da9bb1d2
AC
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);
37f04581 74 debugf3("\tdev = %p\n", mci->dev);
079708b9 75 debugf3("\tmod_name:ctl_name = %s:%s\n", mci->mod_name, mci->ctl_name);
da9bb1d2
AC
76 debugf3("\tpvt_info = %p\n\n", mci->pvt_info);
77}
78
239642fe
BP
79/*
80 * keep those in sync with the enum mem_type
81 */
82const 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};
101EXPORT_SYMBOL_GPL(edac_mem_types);
102
079708b9 103#endif /* CONFIG_EDAC_DEBUG */
da9bb1d2
AC
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 */
7391c6dc 112void *edac_align_ptr(void *ptr, unsigned size)
da9bb1d2
AC
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
079708b9 129 return (char *)ptr;
da9bb1d2
AC
130
131 r = size % align;
132
133 if (r == 0)
079708b9 134 return (char *)ptr;
da9bb1d2 135
7391c6dc 136 return (void *)(((unsigned long)ptr) + align - r);
da9bb1d2
AC
137}
138
da9bb1d2
AC
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 */
155struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows,
b8f6f975 156 unsigned nr_chans, int edac_index)
da9bb1d2
AC
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;
8096cfaf 164 int err;
da9bb1d2
AC
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 */
079708b9 171 mci = (struct mem_ctl_info *)0;
7391c6dc
DT
172 csi = edac_align_ptr(&mci[1], sizeof(*csi));
173 chi = edac_align_ptr(&csi[nr_csrows], sizeof(*chi));
e27e3dac 174 pvt = edac_align_ptr(&chi[nr_chans * nr_csrows], sz_pvt);
079708b9 175 size = ((unsigned long)pvt) + sz_pvt;
da9bb1d2 176
8096cfaf
DT
177 mci = kzalloc(size, GFP_KERNEL);
178 if (mci == NULL)
da9bb1d2
AC
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 */
079708b9
DT
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;
da9bb1d2 187
b8f6f975
DT
188 /* setup index and various internal pointers */
189 mci->mc_idx = edac_index;
da9bb1d2
AC
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
81d87cb1
DJ
209 mci->op_state = OP_ALLOC;
210
8096cfaf
DT
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 */
da9bb1d2
AC
226 return mci;
227}
9110540f 228EXPORT_SYMBOL_GPL(edac_mc_alloc);
da9bb1d2 229
da9bb1d2 230/**
8096cfaf
DT
231 * edac_mc_free
232 * 'Free' a previously allocated 'mci' structure
da9bb1d2 233 * @mci: pointer to a struct mem_ctl_info structure
da9bb1d2
AC
234 */
235void edac_mc_free(struct mem_ctl_info *mci)
236{
8096cfaf 237 edac_mc_unregister_sysfs_main_kobj(mci);
da9bb1d2 238}
9110540f 239EXPORT_SYMBOL_GPL(edac_mc_free);
da9bb1d2 240
bce19683
DT
241
242/*
243 * find_mci_by_dev
244 *
245 * scan list of controllers looking for the one that manages
246 * the 'dev' device
247 */
37f04581 248static struct mem_ctl_info *find_mci_by_dev(struct device *dev)
da9bb1d2
AC
249{
250 struct mem_ctl_info *mci;
251 struct list_head *item;
252
537fba28 253 debugf3("%s()\n", __func__);
da9bb1d2
AC
254
255 list_for_each(item, &mc_devices) {
256 mci = list_entry(item, struct mem_ctl_info, link);
257
37f04581 258 if (mci->dev == dev)
da9bb1d2
AC
259 return mci;
260 }
261
262 return NULL;
263}
264
81d87cb1
DJ
265/*
266 * handler for EDAC to check if NMI type handler has asserted interrupt
267 */
268static int edac_mc_assert_error_check_and_clear(void)
269{
66ee2f94 270 int old_state;
81d87cb1 271
079708b9 272 if (edac_op_state == EDAC_OPSTATE_POLL)
81d87cb1
DJ
273 return 1;
274
66ee2f94
DJ
275 old_state = edac_err_assert;
276 edac_err_assert = 0;
81d87cb1 277
66ee2f94 278 return old_state;
81d87cb1
DJ
279}
280
281/*
282 * edac_mc_workq_function
283 * performs the operation scheduled by a workq request
284 */
81d87cb1
DJ
285static void edac_mc_workq_function(struct work_struct *work_req)
286{
fbeb4384 287 struct delayed_work *d_work = to_delayed_work(work_req);
81d87cb1 288 struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
81d87cb1
DJ
289
290 mutex_lock(&mem_ctls_mutex);
291
bf52fa4a
DT
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
81d87cb1
DJ
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
81d87cb1
DJ
302 mutex_unlock(&mem_ctls_mutex);
303
304 /* Reschedule */
4de78c68 305 queue_delayed_work(edac_workqueue, &mci->work,
052dfb45 306 msecs_to_jiffies(edac_mc_get_poll_msec()));
81d87cb1
DJ
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
bf52fa4a
DT
313 *
314 * locking model:
315 *
316 * called with the mem_ctls_mutex held
81d87cb1 317 */
bf52fa4a 318static void edac_mc_workq_setup(struct mem_ctl_info *mci, unsigned msec)
81d87cb1
DJ
319{
320 debugf0("%s()\n", __func__);
321
bf52fa4a
DT
322 /* if this instance is not in the POLL state, then simply return */
323 if (mci->op_state != OP_RUNNING_POLL)
324 return;
325
81d87cb1 326 INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
81d87cb1
DJ
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
bf52fa4a
DT
333 *
334 * locking model:
335 *
336 * called WITHOUT lock held
81d87cb1 337 */
bf52fa4a 338static void edac_mc_workq_teardown(struct mem_ctl_info *mci)
81d87cb1
DJ
339{
340 int status;
341
bce19683
DT
342 status = cancel_delayed_work(&mci->work);
343 if (status == 0) {
344 debugf0("%s() not canceled, flush the queue\n",
345 __func__);
bf52fa4a 346
bce19683
DT
347 /* workq instance might be running, wait for it */
348 flush_workqueue(edac_workqueue);
81d87cb1
DJ
349 }
350}
351
352/*
bce19683
DT
353 * edac_mc_reset_delay_period(unsigned long value)
354 *
355 * user space has updated our poll period value, need to
356 * reset our workq delays
81d87cb1 357 */
bce19683 358void edac_mc_reset_delay_period(int value)
81d87cb1 359{
bce19683
DT
360 struct mem_ctl_info *mci;
361 struct list_head *item;
362
363 mutex_lock(&mem_ctls_mutex);
364
365 /* scan the list and turn off all workq timers, doing so under lock
366 */
367 list_for_each(item, &mc_devices) {
368 mci = list_entry(item, struct mem_ctl_info, link);
369
370 if (mci->op_state == OP_RUNNING_POLL)
371 cancel_delayed_work(&mci->work);
372 }
373
374 mutex_unlock(&mem_ctls_mutex);
81d87cb1 375
bce19683
DT
376
377 /* re-walk the list, and reset the poll delay */
bf52fa4a
DT
378 mutex_lock(&mem_ctls_mutex);
379
bce19683
DT
380 list_for_each(item, &mc_devices) {
381 mci = list_entry(item, struct mem_ctl_info, link);
382
383 edac_mc_workq_setup(mci, (unsigned long) value);
384 }
81d87cb1
DJ
385
386 mutex_unlock(&mem_ctls_mutex);
387}
388
bce19683
DT
389
390
2d7bbb91
DT
391/* Return 0 on success, 1 on failure.
392 * Before calling this function, caller must
393 * assign a unique value to mci->mc_idx.
bf52fa4a
DT
394 *
395 * locking model:
396 *
397 * called with the mem_ctls_mutex lock held
2d7bbb91 398 */
079708b9 399static int add_mc_to_global_list(struct mem_ctl_info *mci)
da9bb1d2
AC
400{
401 struct list_head *item, *insert_before;
402 struct mem_ctl_info *p;
da9bb1d2 403
2d7bbb91 404 insert_before = &mc_devices;
da9bb1d2 405
bf52fa4a
DT
406 p = find_mci_by_dev(mci->dev);
407 if (unlikely(p != NULL))
2d7bbb91 408 goto fail0;
da9bb1d2 409
2d7bbb91
DT
410 list_for_each(item, &mc_devices) {
411 p = list_entry(item, struct mem_ctl_info, link);
da9bb1d2 412
2d7bbb91
DT
413 if (p->mc_idx >= mci->mc_idx) {
414 if (unlikely(p->mc_idx == mci->mc_idx))
415 goto fail1;
da9bb1d2 416
2d7bbb91
DT
417 insert_before = item;
418 break;
da9bb1d2 419 }
da9bb1d2
AC
420 }
421
422 list_add_tail_rcu(&mci->link, insert_before);
c0d12172 423 atomic_inc(&edac_handlers);
da9bb1d2 424 return 0;
2d7bbb91 425
052dfb45 426fail0:
2d7bbb91 427 edac_printk(KERN_WARNING, EDAC_MC,
281efb17 428 "%s (%s) %s %s already assigned %d\n", dev_name(p->dev),
17aa7e03 429 edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
2d7bbb91
DT
430 return 1;
431
052dfb45 432fail1:
2d7bbb91 433 edac_printk(KERN_WARNING, EDAC_MC,
052dfb45
DT
434 "bug in low-level driver: attempt to assign\n"
435 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
2d7bbb91 436 return 1;
da9bb1d2
AC
437}
438
e7ecd891 439static void complete_mc_list_del(struct rcu_head *head)
a1d03fcc
DP
440{
441 struct mem_ctl_info *mci;
442
443 mci = container_of(head, struct mem_ctl_info, rcu);
444 INIT_LIST_HEAD(&mci->link);
a1d03fcc
DP
445}
446
e7ecd891 447static void del_mc_from_global_list(struct mem_ctl_info *mci)
a1d03fcc 448{
c0d12172 449 atomic_dec(&edac_handlers);
a1d03fcc 450 list_del_rcu(&mci->link);
a1d03fcc 451 call_rcu(&mci->rcu, complete_mc_list_del);
458e5ff1 452 rcu_barrier();
a1d03fcc
DP
453}
454
5da0831c
DT
455/**
456 * edac_mc_find: Search for a mem_ctl_info structure whose index is 'idx'.
457 *
458 * If found, return a pointer to the structure.
459 * Else return NULL.
460 *
461 * Caller must hold mem_ctls_mutex.
462 */
079708b9 463struct mem_ctl_info *edac_mc_find(int idx)
5da0831c
DT
464{
465 struct list_head *item;
466 struct mem_ctl_info *mci;
467
468 list_for_each(item, &mc_devices) {
469 mci = list_entry(item, struct mem_ctl_info, link);
470
471 if (mci->mc_idx >= idx) {
472 if (mci->mc_idx == idx)
473 return mci;
474
475 break;
476 }
477 }
478
479 return NULL;
480}
481EXPORT_SYMBOL(edac_mc_find);
482
da9bb1d2 483/**
472678eb
DP
484 * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and
485 * create sysfs entries associated with mci structure
da9bb1d2 486 * @mci: pointer to the mci structure to be added to the list
2d7bbb91 487 * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure.
da9bb1d2
AC
488 *
489 * Return:
490 * 0 Success
491 * !0 Failure
492 */
493
494/* FIXME - should a warning be printed if no error detection? correction? */
b8f6f975 495int edac_mc_add_mc(struct mem_ctl_info *mci)
da9bb1d2 496{
537fba28 497 debugf0("%s()\n", __func__);
b8f6f975 498
da9bb1d2
AC
499#ifdef CONFIG_EDAC_DEBUG
500 if (edac_debug_level >= 3)
501 edac_mc_dump_mci(mci);
e7ecd891 502
da9bb1d2
AC
503 if (edac_debug_level >= 4) {
504 int i;
505
506 for (i = 0; i < mci->nr_csrows; i++) {
507 int j;
e7ecd891 508
da9bb1d2
AC
509 edac_mc_dump_csrow(&mci->csrows[i]);
510 for (j = 0; j < mci->csrows[i].nr_channels; j++)
079708b9 511 edac_mc_dump_channel(&mci->csrows[i].
052dfb45 512 channels[j]);
da9bb1d2
AC
513 }
514 }
515#endif
63b7df91 516 mutex_lock(&mem_ctls_mutex);
da9bb1d2
AC
517
518 if (add_mc_to_global_list(mci))
028a7b6d 519 goto fail0;
da9bb1d2
AC
520
521 /* set load time so that error rate can be tracked */
522 mci->start_time = jiffies;
523
9794f33d 524 if (edac_create_sysfs_mci_device(mci)) {
525 edac_mc_printk(mci, KERN_WARNING,
052dfb45 526 "failed to create sysfs device\n");
9794f33d 527 goto fail1;
528 }
da9bb1d2 529
81d87cb1
DJ
530 /* If there IS a check routine, then we are running POLLED */
531 if (mci->edac_check != NULL) {
532 /* This instance is NOW RUNNING */
533 mci->op_state = OP_RUNNING_POLL;
534
535 edac_mc_workq_setup(mci, edac_mc_get_poll_msec());
536 } else {
537 mci->op_state = OP_RUNNING_INTERRUPT;
538 }
539
da9bb1d2 540 /* Report action taken */
bf52fa4a 541 edac_mc_printk(mci, KERN_INFO, "Giving out device to '%s' '%s':"
17aa7e03 542 " DEV %s\n", mci->mod_name, mci->ctl_name, edac_dev_name(mci));
da9bb1d2 543
63b7df91 544 mutex_unlock(&mem_ctls_mutex);
028a7b6d 545 return 0;
da9bb1d2 546
052dfb45 547fail1:
028a7b6d
DP
548 del_mc_from_global_list(mci);
549
052dfb45 550fail0:
63b7df91 551 mutex_unlock(&mem_ctls_mutex);
028a7b6d 552 return 1;
da9bb1d2 553}
9110540f 554EXPORT_SYMBOL_GPL(edac_mc_add_mc);
da9bb1d2 555
da9bb1d2 556/**
472678eb
DP
557 * edac_mc_del_mc: Remove sysfs entries for specified mci structure and
558 * remove mci structure from global list
37f04581 559 * @pdev: Pointer to 'struct device' representing mci structure to remove.
da9bb1d2 560 *
18dbc337 561 * Return pointer to removed mci structure, or NULL if device not found.
da9bb1d2 562 */
079708b9 563struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
da9bb1d2 564{
18dbc337 565 struct mem_ctl_info *mci;
da9bb1d2 566
bf52fa4a
DT
567 debugf0("%s()\n", __func__);
568
63b7df91 569 mutex_lock(&mem_ctls_mutex);
18dbc337 570
bf52fa4a
DT
571 /* find the requested mci struct in the global list */
572 mci = find_mci_by_dev(dev);
573 if (mci == NULL) {
63b7df91 574 mutex_unlock(&mem_ctls_mutex);
18dbc337
DP
575 return NULL;
576 }
577
81d87cb1
DJ
578 /* marking MCI offline */
579 mci->op_state = OP_OFFLINE;
580
da9bb1d2 581 del_mc_from_global_list(mci);
63b7df91 582 mutex_unlock(&mem_ctls_mutex);
bf52fa4a
DT
583
584 /* flush workq processes and remove sysfs */
585 edac_mc_workq_teardown(mci);
586 edac_remove_sysfs_mci_device(mci);
587
537fba28 588 edac_printk(KERN_INFO, EDAC_MC,
052dfb45 589 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
17aa7e03 590 mci->mod_name, mci->ctl_name, edac_dev_name(mci));
bf52fa4a 591
18dbc337 592 return mci;
da9bb1d2 593}
9110540f 594EXPORT_SYMBOL_GPL(edac_mc_del_mc);
da9bb1d2 595
2da1c119
AB
596static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
597 u32 size)
da9bb1d2
AC
598{
599 struct page *pg;
600 void *virt_addr;
601 unsigned long flags = 0;
602
537fba28 603 debugf3("%s()\n", __func__);
da9bb1d2
AC
604
605 /* ECC error page was not in our memory. Ignore it. */
079708b9 606 if (!pfn_valid(page))
da9bb1d2
AC
607 return;
608
609 /* Find the actual page structure then map it and fix */
610 pg = pfn_to_page(page);
611
612 if (PageHighMem(pg))
613 local_irq_save(flags);
614
615 virt_addr = kmap_atomic(pg, KM_BOUNCE_READ);
616
617 /* Perform architecture specific atomic scrub operation */
618 atomic_scrub(virt_addr + offset, size);
619
620 /* Unmap and complete */
621 kunmap_atomic(virt_addr, KM_BOUNCE_READ);
622
623 if (PageHighMem(pg))
624 local_irq_restore(flags);
625}
626
da9bb1d2 627/* FIXME - should return -1 */
e7ecd891 628int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
da9bb1d2
AC
629{
630 struct csrow_info *csrows = mci->csrows;
631 int row, i;
632
537fba28 633 debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page);
da9bb1d2
AC
634 row = -1;
635
636 for (i = 0; i < mci->nr_csrows; i++) {
637 struct csrow_info *csrow = &csrows[i];
638
639 if (csrow->nr_pages == 0)
640 continue;
641
537fba28
DP
642 debugf3("MC%d: %s(): first(0x%lx) page(0x%lx) last(0x%lx) "
643 "mask(0x%lx)\n", mci->mc_idx, __func__,
644 csrow->first_page, page, csrow->last_page,
645 csrow->page_mask);
da9bb1d2
AC
646
647 if ((page >= csrow->first_page) &&
648 (page <= csrow->last_page) &&
649 ((page & csrow->page_mask) ==
650 (csrow->first_page & csrow->page_mask))) {
651 row = i;
652 break;
653 }
654 }
655
656 if (row == -1)
537fba28 657 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
658 "could not look up page error address %lx\n",
659 (unsigned long)page);
da9bb1d2
AC
660
661 return row;
662}
9110540f 663EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
da9bb1d2 664
da9bb1d2
AC
665/* FIXME - setable log (warning/emerg) levels */
666/* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */
667void edac_mc_handle_ce(struct mem_ctl_info *mci,
052dfb45
DT
668 unsigned long page_frame_number,
669 unsigned long offset_in_page, unsigned long syndrome,
670 int row, int channel, const char *msg)
da9bb1d2
AC
671{
672 unsigned long remapped_page;
673
537fba28 674 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
da9bb1d2
AC
675
676 /* FIXME - maybe make panic on INTERNAL ERROR an option */
677 if (row >= mci->nr_csrows || row < 0) {
678 /* something is wrong */
537fba28 679 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
680 "INTERNAL ERROR: row out of range "
681 "(%d >= %d)\n", row, mci->nr_csrows);
da9bb1d2
AC
682 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
683 return;
684 }
e7ecd891 685
da9bb1d2
AC
686 if (channel >= mci->csrows[row].nr_channels || channel < 0) {
687 /* something is wrong */
537fba28 688 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
689 "INTERNAL ERROR: channel out of range "
690 "(%d >= %d)\n", channel,
691 mci->csrows[row].nr_channels);
da9bb1d2
AC
692 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
693 return;
694 }
695
4de78c68 696 if (edac_mc_get_log_ce())
da9bb1d2 697 /* FIXME - put in DIMM location */
537fba28 698 edac_mc_printk(mci, KERN_WARNING,
052dfb45
DT
699 "CE page 0x%lx, offset 0x%lx, grain %d, syndrome "
700 "0x%lx, row %d, channel %d, label \"%s\": %s\n",
701 page_frame_number, offset_in_page,
702 mci->csrows[row].grain, syndrome, row, channel,
703 mci->csrows[row].channels[channel].label, msg);
da9bb1d2
AC
704
705 mci->ce_count++;
706 mci->csrows[row].ce_count++;
707 mci->csrows[row].channels[channel].ce_count++;
708
709 if (mci->scrub_mode & SCRUB_SW_SRC) {
710 /*
711 * Some MC's can remap memory so that it is still available
712 * at a different address when PCI devices map into memory.
713 * MC's that can't do this lose the memory where PCI devices
714 * are mapped. This mapping is MC dependant and so we call
715 * back into the MC driver for it to map the MC page to
716 * a physical (CPU) page which can then be mapped to a virtual
717 * page - which can then be scrubbed.
718 */
719 remapped_page = mci->ctl_page_to_phys ?
052dfb45
DT
720 mci->ctl_page_to_phys(mci, page_frame_number) :
721 page_frame_number;
da9bb1d2
AC
722
723 edac_mc_scrub_block(remapped_page, offset_in_page,
052dfb45 724 mci->csrows[row].grain);
da9bb1d2
AC
725 }
726}
9110540f 727EXPORT_SYMBOL_GPL(edac_mc_handle_ce);
da9bb1d2 728
e7ecd891 729void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg)
da9bb1d2 730{
4de78c68 731 if (edac_mc_get_log_ce())
537fba28 732 edac_mc_printk(mci, KERN_WARNING,
052dfb45 733 "CE - no information available: %s\n", msg);
e7ecd891 734
da9bb1d2
AC
735 mci->ce_noinfo_count++;
736 mci->ce_count++;
737}
9110540f 738EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info);
da9bb1d2 739
da9bb1d2 740void edac_mc_handle_ue(struct mem_ctl_info *mci,
052dfb45
DT
741 unsigned long page_frame_number,
742 unsigned long offset_in_page, int row, const char *msg)
da9bb1d2
AC
743{
744 int len = EDAC_MC_LABEL_LEN * 4;
745 char labels[len + 1];
746 char *pos = labels;
747 int chan;
748 int chars;
749
537fba28 750 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
da9bb1d2
AC
751
752 /* FIXME - maybe make panic on INTERNAL ERROR an option */
753 if (row >= mci->nr_csrows || row < 0) {
754 /* something is wrong */
537fba28 755 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
756 "INTERNAL ERROR: row out of range "
757 "(%d >= %d)\n", row, mci->nr_csrows);
da9bb1d2
AC
758 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
759 return;
760 }
761
762 chars = snprintf(pos, len + 1, "%s",
079708b9 763 mci->csrows[row].channels[0].label);
da9bb1d2
AC
764 len -= chars;
765 pos += chars;
e7ecd891 766
da9bb1d2 767 for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0);
052dfb45 768 chan++) {
da9bb1d2 769 chars = snprintf(pos, len + 1, ":%s",
079708b9 770 mci->csrows[row].channels[chan].label);
da9bb1d2
AC
771 len -= chars;
772 pos += chars;
773 }
774
4de78c68 775 if (edac_mc_get_log_ue())
537fba28 776 edac_mc_printk(mci, KERN_EMERG,
052dfb45
DT
777 "UE page 0x%lx, offset 0x%lx, grain %d, row %d, "
778 "labels \"%s\": %s\n", page_frame_number,
779 offset_in_page, mci->csrows[row].grain, row,
780 labels, msg);
da9bb1d2 781
4de78c68 782 if (edac_mc_get_panic_on_ue())
e7ecd891 783 panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, "
052dfb45
DT
784 "row %d, labels \"%s\": %s\n", mci->mc_idx,
785 page_frame_number, offset_in_page,
786 mci->csrows[row].grain, row, labels, msg);
da9bb1d2
AC
787
788 mci->ue_count++;
789 mci->csrows[row].ue_count++;
790}
9110540f 791EXPORT_SYMBOL_GPL(edac_mc_handle_ue);
da9bb1d2 792
e7ecd891 793void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg)
da9bb1d2 794{
4de78c68 795 if (edac_mc_get_panic_on_ue())
da9bb1d2
AC
796 panic("EDAC MC%d: Uncorrected Error", mci->mc_idx);
797
4de78c68 798 if (edac_mc_get_log_ue())
537fba28 799 edac_mc_printk(mci, KERN_WARNING,
052dfb45 800 "UE - no information available: %s\n", msg);
da9bb1d2
AC
801 mci->ue_noinfo_count++;
802 mci->ue_count++;
803}
079708b9 804EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info);
da9bb1d2 805
9794f33d 806/*************************************************************
807 * On Fully Buffered DIMM modules, this help function is
808 * called to process UE events
809 */
810void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci,
052dfb45
DT
811 unsigned int csrow,
812 unsigned int channela,
813 unsigned int channelb, char *msg)
9794f33d 814{
815 int len = EDAC_MC_LABEL_LEN * 4;
816 char labels[len + 1];
817 char *pos = labels;
818 int chars;
819
820 if (csrow >= mci->nr_csrows) {
821 /* something is wrong */
822 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
823 "INTERNAL ERROR: row out of range (%d >= %d)\n",
824 csrow, mci->nr_csrows);
9794f33d 825 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
826 return;
827 }
828
829 if (channela >= mci->csrows[csrow].nr_channels) {
830 /* something is wrong */
831 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
832 "INTERNAL ERROR: channel-a out of range "
833 "(%d >= %d)\n",
834 channela, mci->csrows[csrow].nr_channels);
9794f33d 835 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
836 return;
837 }
838
839 if (channelb >= mci->csrows[csrow].nr_channels) {
840 /* something is wrong */
841 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
842 "INTERNAL ERROR: channel-b out of range "
843 "(%d >= %d)\n",
844 channelb, mci->csrows[csrow].nr_channels);
9794f33d 845 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
846 return;
847 }
848
849 mci->ue_count++;
850 mci->csrows[csrow].ue_count++;
851
852 /* Generate the DIMM labels from the specified channels */
853 chars = snprintf(pos, len + 1, "%s",
854 mci->csrows[csrow].channels[channela].label);
079708b9
DT
855 len -= chars;
856 pos += chars;
9794f33d 857 chars = snprintf(pos, len + 1, "-%s",
858 mci->csrows[csrow].channels[channelb].label);
859
4de78c68 860 if (edac_mc_get_log_ue())
9794f33d 861 edac_mc_printk(mci, KERN_EMERG,
052dfb45
DT
862 "UE row %d, channel-a= %d channel-b= %d "
863 "labels \"%s\": %s\n", csrow, channela, channelb,
864 labels, msg);
9794f33d 865
4de78c68 866 if (edac_mc_get_panic_on_ue())
9794f33d 867 panic("UE row %d, channel-a= %d channel-b= %d "
052dfb45
DT
868 "labels \"%s\": %s\n", csrow, channela,
869 channelb, labels, msg);
9794f33d 870}
871EXPORT_SYMBOL(edac_mc_handle_fbd_ue);
872
873/*************************************************************
874 * On Fully Buffered DIMM modules, this help function is
875 * called to process CE events
876 */
877void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci,
052dfb45 878 unsigned int csrow, unsigned int channel, char *msg)
9794f33d 879{
880
881 /* Ensure boundary values */
882 if (csrow >= mci->nr_csrows) {
883 /* something is wrong */
884 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
885 "INTERNAL ERROR: row out of range (%d >= %d)\n",
886 csrow, mci->nr_csrows);
9794f33d 887 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
888 return;
889 }
890 if (channel >= mci->csrows[csrow].nr_channels) {
891 /* something is wrong */
892 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
893 "INTERNAL ERROR: channel out of range (%d >= %d)\n",
894 channel, mci->csrows[csrow].nr_channels);
9794f33d 895 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
896 return;
897 }
898
4de78c68 899 if (edac_mc_get_log_ce())
9794f33d 900 /* FIXME - put in DIMM location */
901 edac_mc_printk(mci, KERN_WARNING,
052dfb45
DT
902 "CE row %d, channel %d, label \"%s\": %s\n",
903 csrow, channel,
904 mci->csrows[csrow].channels[channel].label, msg);
9794f33d 905
906 mci->ce_count++;
907 mci->csrows[csrow].ce_count++;
908 mci->csrows[csrow].channels[channel].ce_count++;
909}
079708b9 910EXPORT_SYMBOL(edac_mc_handle_fbd_ce);