]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/edac/edac_mc.c
amd64_edac: Fix driver module removal
[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
00740c58
BP
342 if (mci->op_state != OP_RUNNING_POLL)
343 return;
344
bce19683
DT
345 status = cancel_delayed_work(&mci->work);
346 if (status == 0) {
347 debugf0("%s() not canceled, flush the queue\n",
348 __func__);
bf52fa4a 349
bce19683
DT
350 /* workq instance might be running, wait for it */
351 flush_workqueue(edac_workqueue);
81d87cb1
DJ
352 }
353}
354
355/*
bce19683
DT
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
81d87cb1 360 */
bce19683 361void edac_mc_reset_delay_period(int value)
81d87cb1 362{
bce19683
DT
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);
81d87cb1 378
bce19683
DT
379
380 /* re-walk the list, and reset the poll delay */
bf52fa4a
DT
381 mutex_lock(&mem_ctls_mutex);
382
bce19683
DT
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 }
81d87cb1
DJ
388
389 mutex_unlock(&mem_ctls_mutex);
390}
391
bce19683
DT
392
393
2d7bbb91
DT
394/* Return 0 on success, 1 on failure.
395 * Before calling this function, caller must
396 * assign a unique value to mci->mc_idx.
bf52fa4a
DT
397 *
398 * locking model:
399 *
400 * called with the mem_ctls_mutex lock held
2d7bbb91 401 */
079708b9 402static int add_mc_to_global_list(struct mem_ctl_info *mci)
da9bb1d2
AC
403{
404 struct list_head *item, *insert_before;
405 struct mem_ctl_info *p;
da9bb1d2 406
2d7bbb91 407 insert_before = &mc_devices;
da9bb1d2 408
bf52fa4a
DT
409 p = find_mci_by_dev(mci->dev);
410 if (unlikely(p != NULL))
2d7bbb91 411 goto fail0;
da9bb1d2 412
2d7bbb91
DT
413 list_for_each(item, &mc_devices) {
414 p = list_entry(item, struct mem_ctl_info, link);
da9bb1d2 415
2d7bbb91
DT
416 if (p->mc_idx >= mci->mc_idx) {
417 if (unlikely(p->mc_idx == mci->mc_idx))
418 goto fail1;
da9bb1d2 419
2d7bbb91
DT
420 insert_before = item;
421 break;
da9bb1d2 422 }
da9bb1d2
AC
423 }
424
425 list_add_tail_rcu(&mci->link, insert_before);
c0d12172 426 atomic_inc(&edac_handlers);
da9bb1d2 427 return 0;
2d7bbb91 428
052dfb45 429fail0:
2d7bbb91 430 edac_printk(KERN_WARNING, EDAC_MC,
281efb17 431 "%s (%s) %s %s already assigned %d\n", dev_name(p->dev),
17aa7e03 432 edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
2d7bbb91
DT
433 return 1;
434
052dfb45 435fail1:
2d7bbb91 436 edac_printk(KERN_WARNING, EDAC_MC,
052dfb45
DT
437 "bug in low-level driver: attempt to assign\n"
438 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
2d7bbb91 439 return 1;
da9bb1d2
AC
440}
441
e7ecd891 442static void complete_mc_list_del(struct rcu_head *head)
a1d03fcc
DP
443{
444 struct mem_ctl_info *mci;
445
446 mci = container_of(head, struct mem_ctl_info, rcu);
447 INIT_LIST_HEAD(&mci->link);
a1d03fcc
DP
448}
449
e7ecd891 450static void del_mc_from_global_list(struct mem_ctl_info *mci)
a1d03fcc 451{
c0d12172 452 atomic_dec(&edac_handlers);
a1d03fcc 453 list_del_rcu(&mci->link);
a1d03fcc 454 call_rcu(&mci->rcu, complete_mc_list_del);
458e5ff1 455 rcu_barrier();
a1d03fcc
DP
456}
457
5da0831c
DT
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 */
079708b9 466struct mem_ctl_info *edac_mc_find(int idx)
5da0831c
DT
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}
484EXPORT_SYMBOL(edac_mc_find);
485
da9bb1d2 486/**
472678eb
DP
487 * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and
488 * create sysfs entries associated with mci structure
da9bb1d2 489 * @mci: pointer to the mci structure to be added to the list
2d7bbb91 490 * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure.
da9bb1d2
AC
491 *
492 * Return:
493 * 0 Success
494 * !0 Failure
495 */
496
497/* FIXME - should a warning be printed if no error detection? correction? */
b8f6f975 498int edac_mc_add_mc(struct mem_ctl_info *mci)
da9bb1d2 499{
537fba28 500 debugf0("%s()\n", __func__);
b8f6f975 501
da9bb1d2
AC
502#ifdef CONFIG_EDAC_DEBUG
503 if (edac_debug_level >= 3)
504 edac_mc_dump_mci(mci);
e7ecd891 505
da9bb1d2
AC
506 if (edac_debug_level >= 4) {
507 int i;
508
509 for (i = 0; i < mci->nr_csrows; i++) {
510 int j;
e7ecd891 511
da9bb1d2
AC
512 edac_mc_dump_csrow(&mci->csrows[i]);
513 for (j = 0; j < mci->csrows[i].nr_channels; j++)
079708b9 514 edac_mc_dump_channel(&mci->csrows[i].
052dfb45 515 channels[j]);
da9bb1d2
AC
516 }
517 }
518#endif
63b7df91 519 mutex_lock(&mem_ctls_mutex);
da9bb1d2
AC
520
521 if (add_mc_to_global_list(mci))
028a7b6d 522 goto fail0;
da9bb1d2
AC
523
524 /* set load time so that error rate can be tracked */
525 mci->start_time = jiffies;
526
9794f33d 527 if (edac_create_sysfs_mci_device(mci)) {
528 edac_mc_printk(mci, KERN_WARNING,
052dfb45 529 "failed to create sysfs device\n");
9794f33d 530 goto fail1;
531 }
da9bb1d2 532
81d87cb1
DJ
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
da9bb1d2 543 /* Report action taken */
bf52fa4a 544 edac_mc_printk(mci, KERN_INFO, "Giving out device to '%s' '%s':"
17aa7e03 545 " DEV %s\n", mci->mod_name, mci->ctl_name, edac_dev_name(mci));
da9bb1d2 546
63b7df91 547 mutex_unlock(&mem_ctls_mutex);
028a7b6d 548 return 0;
da9bb1d2 549
052dfb45 550fail1:
028a7b6d
DP
551 del_mc_from_global_list(mci);
552
052dfb45 553fail0:
63b7df91 554 mutex_unlock(&mem_ctls_mutex);
028a7b6d 555 return 1;
da9bb1d2 556}
9110540f 557EXPORT_SYMBOL_GPL(edac_mc_add_mc);
da9bb1d2 558
da9bb1d2 559/**
472678eb
DP
560 * edac_mc_del_mc: Remove sysfs entries for specified mci structure and
561 * remove mci structure from global list
37f04581 562 * @pdev: Pointer to 'struct device' representing mci structure to remove.
da9bb1d2 563 *
18dbc337 564 * Return pointer to removed mci structure, or NULL if device not found.
da9bb1d2 565 */
079708b9 566struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
da9bb1d2 567{
18dbc337 568 struct mem_ctl_info *mci;
da9bb1d2 569
bf52fa4a
DT
570 debugf0("%s()\n", __func__);
571
63b7df91 572 mutex_lock(&mem_ctls_mutex);
18dbc337 573
bf52fa4a
DT
574 /* find the requested mci struct in the global list */
575 mci = find_mci_by_dev(dev);
576 if (mci == NULL) {
63b7df91 577 mutex_unlock(&mem_ctls_mutex);
18dbc337
DP
578 return NULL;
579 }
580
81d87cb1
DJ
581 /* marking MCI offline */
582 mci->op_state = OP_OFFLINE;
583
da9bb1d2 584 del_mc_from_global_list(mci);
63b7df91 585 mutex_unlock(&mem_ctls_mutex);
bf52fa4a
DT
586
587 /* flush workq processes and remove sysfs */
588 edac_mc_workq_teardown(mci);
589 edac_remove_sysfs_mci_device(mci);
590
537fba28 591 edac_printk(KERN_INFO, EDAC_MC,
052dfb45 592 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
17aa7e03 593 mci->mod_name, mci->ctl_name, edac_dev_name(mci));
bf52fa4a 594
18dbc337 595 return mci;
da9bb1d2 596}
9110540f 597EXPORT_SYMBOL_GPL(edac_mc_del_mc);
da9bb1d2 598
2da1c119
AB
599static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
600 u32 size)
da9bb1d2
AC
601{
602 struct page *pg;
603 void *virt_addr;
604 unsigned long flags = 0;
605
537fba28 606 debugf3("%s()\n", __func__);
da9bb1d2
AC
607
608 /* ECC error page was not in our memory. Ignore it. */
079708b9 609 if (!pfn_valid(page))
da9bb1d2
AC
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
da9bb1d2 630/* FIXME - should return -1 */
e7ecd891 631int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
da9bb1d2
AC
632{
633 struct csrow_info *csrows = mci->csrows;
634 int row, i;
635
537fba28 636 debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page);
da9bb1d2
AC
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
537fba28
DP
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);
da9bb1d2
AC
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)
537fba28 660 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
661 "could not look up page error address %lx\n",
662 (unsigned long)page);
da9bb1d2
AC
663
664 return row;
665}
9110540f 666EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
da9bb1d2 667
da9bb1d2
AC
668/* FIXME - setable log (warning/emerg) levels */
669/* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */
670void edac_mc_handle_ce(struct mem_ctl_info *mci,
052dfb45
DT
671 unsigned long page_frame_number,
672 unsigned long offset_in_page, unsigned long syndrome,
673 int row, int channel, const char *msg)
da9bb1d2
AC
674{
675 unsigned long remapped_page;
676
537fba28 677 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
da9bb1d2
AC
678
679 /* FIXME - maybe make panic on INTERNAL ERROR an option */
680 if (row >= mci->nr_csrows || row < 0) {
681 /* something is wrong */
537fba28 682 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
683 "INTERNAL ERROR: row out of range "
684 "(%d >= %d)\n", row, mci->nr_csrows);
da9bb1d2
AC
685 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
686 return;
687 }
e7ecd891 688
da9bb1d2
AC
689 if (channel >= mci->csrows[row].nr_channels || channel < 0) {
690 /* something is wrong */
537fba28 691 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
692 "INTERNAL ERROR: channel out of range "
693 "(%d >= %d)\n", channel,
694 mci->csrows[row].nr_channels);
da9bb1d2
AC
695 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
696 return;
697 }
698
4de78c68 699 if (edac_mc_get_log_ce())
da9bb1d2 700 /* FIXME - put in DIMM location */
537fba28 701 edac_mc_printk(mci, KERN_WARNING,
052dfb45
DT
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);
da9bb1d2
AC
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 ?
052dfb45
DT
723 mci->ctl_page_to_phys(mci, page_frame_number) :
724 page_frame_number;
da9bb1d2
AC
725
726 edac_mc_scrub_block(remapped_page, offset_in_page,
052dfb45 727 mci->csrows[row].grain);
da9bb1d2
AC
728 }
729}
9110540f 730EXPORT_SYMBOL_GPL(edac_mc_handle_ce);
da9bb1d2 731
e7ecd891 732void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg)
da9bb1d2 733{
4de78c68 734 if (edac_mc_get_log_ce())
537fba28 735 edac_mc_printk(mci, KERN_WARNING,
052dfb45 736 "CE - no information available: %s\n", msg);
e7ecd891 737
da9bb1d2
AC
738 mci->ce_noinfo_count++;
739 mci->ce_count++;
740}
9110540f 741EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info);
da9bb1d2 742
da9bb1d2 743void edac_mc_handle_ue(struct mem_ctl_info *mci,
052dfb45
DT
744 unsigned long page_frame_number,
745 unsigned long offset_in_page, int row, const char *msg)
da9bb1d2
AC
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
537fba28 753 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
da9bb1d2
AC
754
755 /* FIXME - maybe make panic on INTERNAL ERROR an option */
756 if (row >= mci->nr_csrows || row < 0) {
757 /* something is wrong */
537fba28 758 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
759 "INTERNAL ERROR: row out of range "
760 "(%d >= %d)\n", row, mci->nr_csrows);
da9bb1d2
AC
761 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
762 return;
763 }
764
765 chars = snprintf(pos, len + 1, "%s",
079708b9 766 mci->csrows[row].channels[0].label);
da9bb1d2
AC
767 len -= chars;
768 pos += chars;
e7ecd891 769
da9bb1d2 770 for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0);
052dfb45 771 chan++) {
da9bb1d2 772 chars = snprintf(pos, len + 1, ":%s",
079708b9 773 mci->csrows[row].channels[chan].label);
da9bb1d2
AC
774 len -= chars;
775 pos += chars;
776 }
777
4de78c68 778 if (edac_mc_get_log_ue())
537fba28 779 edac_mc_printk(mci, KERN_EMERG,
052dfb45
DT
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);
da9bb1d2 784
4de78c68 785 if (edac_mc_get_panic_on_ue())
e7ecd891 786 panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, "
052dfb45
DT
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);
da9bb1d2
AC
790
791 mci->ue_count++;
792 mci->csrows[row].ue_count++;
793}
9110540f 794EXPORT_SYMBOL_GPL(edac_mc_handle_ue);
da9bb1d2 795
e7ecd891 796void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg)
da9bb1d2 797{
4de78c68 798 if (edac_mc_get_panic_on_ue())
da9bb1d2
AC
799 panic("EDAC MC%d: Uncorrected Error", mci->mc_idx);
800
4de78c68 801 if (edac_mc_get_log_ue())
537fba28 802 edac_mc_printk(mci, KERN_WARNING,
052dfb45 803 "UE - no information available: %s\n", msg);
da9bb1d2
AC
804 mci->ue_noinfo_count++;
805 mci->ue_count++;
806}
079708b9 807EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info);
da9bb1d2 808
9794f33d 809/*************************************************************
810 * On Fully Buffered DIMM modules, this help function is
811 * called to process UE events
812 */
813void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci,
052dfb45
DT
814 unsigned int csrow,
815 unsigned int channela,
816 unsigned int channelb, char *msg)
9794f33d 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,
052dfb45
DT
826 "INTERNAL ERROR: row out of range (%d >= %d)\n",
827 csrow, mci->nr_csrows);
9794f33d 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,
052dfb45
DT
835 "INTERNAL ERROR: channel-a out of range "
836 "(%d >= %d)\n",
837 channela, mci->csrows[csrow].nr_channels);
9794f33d 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,
052dfb45
DT
845 "INTERNAL ERROR: channel-b out of range "
846 "(%d >= %d)\n",
847 channelb, mci->csrows[csrow].nr_channels);
9794f33d 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);
079708b9
DT
858 len -= chars;
859 pos += chars;
9794f33d 860 chars = snprintf(pos, len + 1, "-%s",
861 mci->csrows[csrow].channels[channelb].label);
862
4de78c68 863 if (edac_mc_get_log_ue())
9794f33d 864 edac_mc_printk(mci, KERN_EMERG,
052dfb45
DT
865 "UE row %d, channel-a= %d channel-b= %d "
866 "labels \"%s\": %s\n", csrow, channela, channelb,
867 labels, msg);
9794f33d 868
4de78c68 869 if (edac_mc_get_panic_on_ue())
9794f33d 870 panic("UE row %d, channel-a= %d channel-b= %d "
052dfb45
DT
871 "labels \"%s\": %s\n", csrow, channela,
872 channelb, labels, msg);
9794f33d 873}
874EXPORT_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 */
880void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci,
052dfb45 881 unsigned int csrow, unsigned int channel, char *msg)
9794f33d 882{
883
884 /* Ensure boundary values */
885 if (csrow >= mci->nr_csrows) {
886 /* something is wrong */
887 edac_mc_printk(mci, KERN_ERR,
052dfb45
DT
888 "INTERNAL ERROR: row out of range (%d >= %d)\n",
889 csrow, mci->nr_csrows);
9794f33d 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,
052dfb45
DT
896 "INTERNAL ERROR: channel out of range (%d >= %d)\n",
897 channel, mci->csrows[csrow].nr_channels);
9794f33d 898 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
899 return;
900 }
901
4de78c68 902 if (edac_mc_get_log_ce())
9794f33d 903 /* FIXME - put in DIMM location */
904 edac_mc_printk(mci, KERN_WARNING,
052dfb45
DT
905 "CE row %d, channel %d, label \"%s\": %s\n",
906 csrow, channel,
907 mci->csrows[csrow].channels[channel].label, msg);
9794f33d 908
909 mci->ce_count++;
910 mci->csrows[csrow].ce_count++;
911 mci->csrows[csrow].channels[channel].ce_count++;
912}
079708b9 913EXPORT_SYMBOL(edac_mc_handle_fbd_ce);