ALSA: usb-audio: add mapping for ASRock TRX40 Creator
[linux/fpc-iii.git] / drivers / edac / edac_mc.c
blobfd440b35d76ed29bd74e158fc7a70935627fb00d
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.
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/
11 * Modified by Dave Peterson and Doug Thompson
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/ctype.h>
29 #include <linux/edac.h>
30 #include <linux/bitops.h>
31 #include <linux/uaccess.h>
32 #include <asm/page.h>
33 #include "edac_mc.h"
34 #include "edac_module.h"
35 #include <ras/ras_event.h>
37 #ifdef CONFIG_EDAC_ATOMIC_SCRUB
38 #include <asm/edac.h>
39 #else
40 #define edac_atomic_scrub(va, size) do { } while (0)
41 #endif
43 int edac_op_state = EDAC_OPSTATE_INVAL;
44 EXPORT_SYMBOL_GPL(edac_op_state);
46 static int edac_report = EDAC_REPORTING_ENABLED;
48 /* lock to memory controller's control array */
49 static DEFINE_MUTEX(mem_ctls_mutex);
50 static LIST_HEAD(mc_devices);
53 * Used to lock EDAC MC to just one module, avoiding two drivers e. g.
54 * apei/ghes and i7core_edac to be used at the same time.
56 static const char *edac_mc_owner;
58 static struct bus_type mc_bus[EDAC_MAX_MCS];
60 int edac_get_report_status(void)
62 return edac_report;
64 EXPORT_SYMBOL_GPL(edac_get_report_status);
66 void edac_set_report_status(int new)
68 if (new == EDAC_REPORTING_ENABLED ||
69 new == EDAC_REPORTING_DISABLED ||
70 new == EDAC_REPORTING_FORCE)
71 edac_report = new;
73 EXPORT_SYMBOL_GPL(edac_set_report_status);
75 static int edac_report_set(const char *str, const struct kernel_param *kp)
77 if (!str)
78 return -EINVAL;
80 if (!strncmp(str, "on", 2))
81 edac_report = EDAC_REPORTING_ENABLED;
82 else if (!strncmp(str, "off", 3))
83 edac_report = EDAC_REPORTING_DISABLED;
84 else if (!strncmp(str, "force", 5))
85 edac_report = EDAC_REPORTING_FORCE;
87 return 0;
90 static int edac_report_get(char *buffer, const struct kernel_param *kp)
92 int ret = 0;
94 switch (edac_report) {
95 case EDAC_REPORTING_ENABLED:
96 ret = sprintf(buffer, "on");
97 break;
98 case EDAC_REPORTING_DISABLED:
99 ret = sprintf(buffer, "off");
100 break;
101 case EDAC_REPORTING_FORCE:
102 ret = sprintf(buffer, "force");
103 break;
104 default:
105 ret = -EINVAL;
106 break;
109 return ret;
112 static const struct kernel_param_ops edac_report_ops = {
113 .set = edac_report_set,
114 .get = edac_report_get,
117 module_param_cb(edac_report, &edac_report_ops, &edac_report, 0644);
119 unsigned edac_dimm_info_location(struct dimm_info *dimm, char *buf,
120 unsigned len)
122 struct mem_ctl_info *mci = dimm->mci;
123 int i, n, count = 0;
124 char *p = buf;
126 for (i = 0; i < mci->n_layers; i++) {
127 n = snprintf(p, len, "%s %d ",
128 edac_layer_name[mci->layers[i].type],
129 dimm->location[i]);
130 p += n;
131 len -= n;
132 count += n;
133 if (!len)
134 break;
137 return count;
140 #ifdef CONFIG_EDAC_DEBUG
142 static void edac_mc_dump_channel(struct rank_info *chan)
144 edac_dbg(4, " channel->chan_idx = %d\n", chan->chan_idx);
145 edac_dbg(4, " channel = %p\n", chan);
146 edac_dbg(4, " channel->csrow = %p\n", chan->csrow);
147 edac_dbg(4, " channel->dimm = %p\n", chan->dimm);
150 static void edac_mc_dump_dimm(struct dimm_info *dimm, int number)
152 char location[80];
154 edac_dimm_info_location(dimm, location, sizeof(location));
156 edac_dbg(4, "%s%i: %smapped as virtual row %d, chan %d\n",
157 dimm->mci->csbased ? "rank" : "dimm",
158 number, location, dimm->csrow, dimm->cschannel);
159 edac_dbg(4, " dimm = %p\n", dimm);
160 edac_dbg(4, " dimm->label = '%s'\n", dimm->label);
161 edac_dbg(4, " dimm->nr_pages = 0x%x\n", dimm->nr_pages);
162 edac_dbg(4, " dimm->grain = %d\n", dimm->grain);
163 edac_dbg(4, " dimm->nr_pages = 0x%x\n", dimm->nr_pages);
166 static void edac_mc_dump_csrow(struct csrow_info *csrow)
168 edac_dbg(4, "csrow->csrow_idx = %d\n", csrow->csrow_idx);
169 edac_dbg(4, " csrow = %p\n", csrow);
170 edac_dbg(4, " csrow->first_page = 0x%lx\n", csrow->first_page);
171 edac_dbg(4, " csrow->last_page = 0x%lx\n", csrow->last_page);
172 edac_dbg(4, " csrow->page_mask = 0x%lx\n", csrow->page_mask);
173 edac_dbg(4, " csrow->nr_channels = %d\n", csrow->nr_channels);
174 edac_dbg(4, " csrow->channels = %p\n", csrow->channels);
175 edac_dbg(4, " csrow->mci = %p\n", csrow->mci);
178 static void edac_mc_dump_mci(struct mem_ctl_info *mci)
180 edac_dbg(3, "\tmci = %p\n", mci);
181 edac_dbg(3, "\tmci->mtype_cap = %lx\n", mci->mtype_cap);
182 edac_dbg(3, "\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap);
183 edac_dbg(3, "\tmci->edac_cap = %lx\n", mci->edac_cap);
184 edac_dbg(4, "\tmci->edac_check = %p\n", mci->edac_check);
185 edac_dbg(3, "\tmci->nr_csrows = %d, csrows = %p\n",
186 mci->nr_csrows, mci->csrows);
187 edac_dbg(3, "\tmci->nr_dimms = %d, dimms = %p\n",
188 mci->tot_dimms, mci->dimms);
189 edac_dbg(3, "\tdev = %p\n", mci->pdev);
190 edac_dbg(3, "\tmod_name:ctl_name = %s:%s\n",
191 mci->mod_name, mci->ctl_name);
192 edac_dbg(3, "\tpvt_info = %p\n\n", mci->pvt_info);
195 #endif /* CONFIG_EDAC_DEBUG */
197 const char * const edac_mem_types[] = {
198 [MEM_EMPTY] = "Empty",
199 [MEM_RESERVED] = "Reserved",
200 [MEM_UNKNOWN] = "Unknown",
201 [MEM_FPM] = "FPM",
202 [MEM_EDO] = "EDO",
203 [MEM_BEDO] = "BEDO",
204 [MEM_SDR] = "Unbuffered-SDR",
205 [MEM_RDR] = "Registered-SDR",
206 [MEM_DDR] = "Unbuffered-DDR",
207 [MEM_RDDR] = "Registered-DDR",
208 [MEM_RMBS] = "RMBS",
209 [MEM_DDR2] = "Unbuffered-DDR2",
210 [MEM_FB_DDR2] = "FullyBuffered-DDR2",
211 [MEM_RDDR2] = "Registered-DDR2",
212 [MEM_XDR] = "XDR",
213 [MEM_DDR3] = "Unbuffered-DDR3",
214 [MEM_RDDR3] = "Registered-DDR3",
215 [MEM_LRDDR3] = "Load-Reduced-DDR3-RAM",
216 [MEM_DDR4] = "Unbuffered-DDR4",
217 [MEM_RDDR4] = "Registered-DDR4",
218 [MEM_LRDDR4] = "Load-Reduced-DDR4-RAM",
219 [MEM_NVDIMM] = "Non-volatile-RAM",
221 EXPORT_SYMBOL_GPL(edac_mem_types);
224 * edac_align_ptr - Prepares the pointer offsets for a single-shot allocation
225 * @p: pointer to a pointer with the memory offset to be used. At
226 * return, this will be incremented to point to the next offset
227 * @size: Size of the data structure to be reserved
228 * @n_elems: Number of elements that should be reserved
230 * If 'size' is a constant, the compiler will optimize this whole function
231 * down to either a no-op or the addition of a constant to the value of '*p'.
233 * The 'p' pointer is absolutely needed to keep the proper advancing
234 * further in memory to the proper offsets when allocating the struct along
235 * with its embedded structs, as edac_device_alloc_ctl_info() does it
236 * above, for example.
238 * At return, the pointer 'p' will be incremented to be used on a next call
239 * to this function.
241 void *edac_align_ptr(void **p, unsigned size, int n_elems)
243 unsigned align, r;
244 void *ptr = *p;
246 *p += size * n_elems;
249 * 'p' can possibly be an unaligned item X such that sizeof(X) is
250 * 'size'. Adjust 'p' so that its alignment is at least as
251 * stringent as what the compiler would provide for X and return
252 * the aligned result.
253 * Here we assume that the alignment of a "long long" is the most
254 * stringent alignment that the compiler will ever provide by default.
255 * As far as I know, this is a reasonable assumption.
257 if (size > sizeof(long))
258 align = sizeof(long long);
259 else if (size > sizeof(int))
260 align = sizeof(long);
261 else if (size > sizeof(short))
262 align = sizeof(int);
263 else if (size > sizeof(char))
264 align = sizeof(short);
265 else
266 return (char *)ptr;
268 r = (unsigned long)p % align;
270 if (r == 0)
271 return (char *)ptr;
273 *p += align - r;
275 return (void *)(((unsigned long)ptr) + align - r);
278 static void _edac_mc_free(struct mem_ctl_info *mci)
280 int i, chn, row;
281 struct csrow_info *csr;
282 const unsigned int tot_dimms = mci->tot_dimms;
283 const unsigned int tot_channels = mci->num_cschannel;
284 const unsigned int tot_csrows = mci->nr_csrows;
286 if (mci->dimms) {
287 for (i = 0; i < tot_dimms; i++)
288 kfree(mci->dimms[i]);
289 kfree(mci->dimms);
291 if (mci->csrows) {
292 for (row = 0; row < tot_csrows; row++) {
293 csr = mci->csrows[row];
294 if (csr) {
295 if (csr->channels) {
296 for (chn = 0; chn < tot_channels; chn++)
297 kfree(csr->channels[chn]);
298 kfree(csr->channels);
300 kfree(csr);
303 kfree(mci->csrows);
305 kfree(mci);
308 struct mem_ctl_info *edac_mc_alloc(unsigned mc_num,
309 unsigned n_layers,
310 struct edac_mc_layer *layers,
311 unsigned sz_pvt)
313 struct mem_ctl_info *mci;
314 struct edac_mc_layer *layer;
315 struct csrow_info *csr;
316 struct rank_info *chan;
317 struct dimm_info *dimm;
318 u32 *ce_per_layer[EDAC_MAX_LAYERS], *ue_per_layer[EDAC_MAX_LAYERS];
319 unsigned pos[EDAC_MAX_LAYERS];
320 unsigned size, tot_dimms = 1, count = 1;
321 unsigned tot_csrows = 1, tot_channels = 1, tot_errcount = 0;
322 void *pvt, *p, *ptr = NULL;
323 int i, j, row, chn, n, len, off;
324 bool per_rank = false;
326 BUG_ON(n_layers > EDAC_MAX_LAYERS || n_layers == 0);
328 * Calculate the total amount of dimms and csrows/cschannels while
329 * in the old API emulation mode
331 for (i = 0; i < n_layers; i++) {
332 tot_dimms *= layers[i].size;
333 if (layers[i].is_virt_csrow)
334 tot_csrows *= layers[i].size;
335 else
336 tot_channels *= layers[i].size;
338 if (layers[i].type == EDAC_MC_LAYER_CHIP_SELECT)
339 per_rank = true;
342 /* Figure out the offsets of the various items from the start of an mc
343 * structure. We want the alignment of each item to be at least as
344 * stringent as what the compiler would provide if we could simply
345 * hardcode everything into a single struct.
347 mci = edac_align_ptr(&ptr, sizeof(*mci), 1);
348 layer = edac_align_ptr(&ptr, sizeof(*layer), n_layers);
349 for (i = 0; i < n_layers; i++) {
350 count *= layers[i].size;
351 edac_dbg(4, "errcount layer %d size %d\n", i, count);
352 ce_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count);
353 ue_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count);
354 tot_errcount += 2 * count;
357 edac_dbg(4, "allocating %d error counters\n", tot_errcount);
358 pvt = edac_align_ptr(&ptr, sz_pvt, 1);
359 size = ((unsigned long)pvt) + sz_pvt;
361 edac_dbg(1, "allocating %u bytes for mci data (%d %s, %d csrows/channels)\n",
362 size,
363 tot_dimms,
364 per_rank ? "ranks" : "dimms",
365 tot_csrows * tot_channels);
367 mci = kzalloc(size, GFP_KERNEL);
368 if (mci == NULL)
369 return NULL;
371 /* Adjust pointers so they point within the memory we just allocated
372 * rather than an imaginary chunk of memory located at address 0.
374 layer = (struct edac_mc_layer *)(((char *)mci) + ((unsigned long)layer));
375 for (i = 0; i < n_layers; i++) {
376 mci->ce_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ce_per_layer[i]));
377 mci->ue_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ue_per_layer[i]));
379 pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL;
381 /* setup index and various internal pointers */
382 mci->mc_idx = mc_num;
383 mci->tot_dimms = tot_dimms;
384 mci->pvt_info = pvt;
385 mci->n_layers = n_layers;
386 mci->layers = layer;
387 memcpy(mci->layers, layers, sizeof(*layer) * n_layers);
388 mci->nr_csrows = tot_csrows;
389 mci->num_cschannel = tot_channels;
390 mci->csbased = per_rank;
393 * Alocate and fill the csrow/channels structs
395 mci->csrows = kcalloc(tot_csrows, sizeof(*mci->csrows), GFP_KERNEL);
396 if (!mci->csrows)
397 goto error;
398 for (row = 0; row < tot_csrows; row++) {
399 csr = kzalloc(sizeof(**mci->csrows), GFP_KERNEL);
400 if (!csr)
401 goto error;
402 mci->csrows[row] = csr;
403 csr->csrow_idx = row;
404 csr->mci = mci;
405 csr->nr_channels = tot_channels;
406 csr->channels = kcalloc(tot_channels, sizeof(*csr->channels),
407 GFP_KERNEL);
408 if (!csr->channels)
409 goto error;
411 for (chn = 0; chn < tot_channels; chn++) {
412 chan = kzalloc(sizeof(**csr->channels), GFP_KERNEL);
413 if (!chan)
414 goto error;
415 csr->channels[chn] = chan;
416 chan->chan_idx = chn;
417 chan->csrow = csr;
422 * Allocate and fill the dimm structs
424 mci->dimms = kcalloc(tot_dimms, sizeof(*mci->dimms), GFP_KERNEL);
425 if (!mci->dimms)
426 goto error;
428 memset(&pos, 0, sizeof(pos));
429 row = 0;
430 chn = 0;
431 for (i = 0; i < tot_dimms; i++) {
432 chan = mci->csrows[row]->channels[chn];
433 off = EDAC_DIMM_OFF(layer, n_layers, pos[0], pos[1], pos[2]);
434 if (off < 0 || off >= tot_dimms) {
435 edac_mc_printk(mci, KERN_ERR, "EDAC core bug: EDAC_DIMM_OFF is trying to do an illegal data access\n");
436 goto error;
439 dimm = kzalloc(sizeof(**mci->dimms), GFP_KERNEL);
440 if (!dimm)
441 goto error;
442 mci->dimms[off] = dimm;
443 dimm->mci = mci;
446 * Copy DIMM location and initialize it.
448 len = sizeof(dimm->label);
449 p = dimm->label;
450 n = snprintf(p, len, "mc#%u", mc_num);
451 p += n;
452 len -= n;
453 for (j = 0; j < n_layers; j++) {
454 n = snprintf(p, len, "%s#%u",
455 edac_layer_name[layers[j].type],
456 pos[j]);
457 p += n;
458 len -= n;
459 dimm->location[j] = pos[j];
461 if (len <= 0)
462 break;
465 /* Link it to the csrows old API data */
466 chan->dimm = dimm;
467 dimm->csrow = row;
468 dimm->cschannel = chn;
470 /* Increment csrow location */
471 if (layers[0].is_virt_csrow) {
472 chn++;
473 if (chn == tot_channels) {
474 chn = 0;
475 row++;
477 } else {
478 row++;
479 if (row == tot_csrows) {
480 row = 0;
481 chn++;
485 /* Increment dimm location */
486 for (j = n_layers - 1; j >= 0; j--) {
487 pos[j]++;
488 if (pos[j] < layers[j].size)
489 break;
490 pos[j] = 0;
494 mci->op_state = OP_ALLOC;
496 return mci;
498 error:
499 _edac_mc_free(mci);
501 return NULL;
503 EXPORT_SYMBOL_GPL(edac_mc_alloc);
505 void edac_mc_free(struct mem_ctl_info *mci)
507 edac_dbg(1, "\n");
509 /* If we're not yet registered with sysfs free only what was allocated
510 * in edac_mc_alloc().
512 if (!device_is_registered(&mci->dev)) {
513 _edac_mc_free(mci);
514 return;
517 /* the mci instance is freed here, when the sysfs object is dropped */
518 edac_unregister_sysfs(mci);
520 EXPORT_SYMBOL_GPL(edac_mc_free);
522 bool edac_has_mcs(void)
524 bool ret;
526 mutex_lock(&mem_ctls_mutex);
528 ret = list_empty(&mc_devices);
530 mutex_unlock(&mem_ctls_mutex);
532 return !ret;
534 EXPORT_SYMBOL_GPL(edac_has_mcs);
536 /* Caller must hold mem_ctls_mutex */
537 static struct mem_ctl_info *__find_mci_by_dev(struct device *dev)
539 struct mem_ctl_info *mci;
540 struct list_head *item;
542 edac_dbg(3, "\n");
544 list_for_each(item, &mc_devices) {
545 mci = list_entry(item, struct mem_ctl_info, link);
547 if (mci->pdev == dev)
548 return mci;
551 return NULL;
555 * find_mci_by_dev
557 * scan list of controllers looking for the one that manages
558 * the 'dev' device
559 * @dev: pointer to a struct device related with the MCI
561 struct mem_ctl_info *find_mci_by_dev(struct device *dev)
563 struct mem_ctl_info *ret;
565 mutex_lock(&mem_ctls_mutex);
566 ret = __find_mci_by_dev(dev);
567 mutex_unlock(&mem_ctls_mutex);
569 return ret;
571 EXPORT_SYMBOL_GPL(find_mci_by_dev);
574 * edac_mc_workq_function
575 * performs the operation scheduled by a workq request
577 static void edac_mc_workq_function(struct work_struct *work_req)
579 struct delayed_work *d_work = to_delayed_work(work_req);
580 struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
582 mutex_lock(&mem_ctls_mutex);
584 if (mci->op_state != OP_RUNNING_POLL) {
585 mutex_unlock(&mem_ctls_mutex);
586 return;
589 if (edac_op_state == EDAC_OPSTATE_POLL)
590 mci->edac_check(mci);
592 mutex_unlock(&mem_ctls_mutex);
594 /* Queue ourselves again. */
595 edac_queue_work(&mci->work, msecs_to_jiffies(edac_mc_get_poll_msec()));
599 * edac_mc_reset_delay_period(unsigned long value)
601 * user space has updated our poll period value, need to
602 * reset our workq delays
604 void edac_mc_reset_delay_period(unsigned long value)
606 struct mem_ctl_info *mci;
607 struct list_head *item;
609 mutex_lock(&mem_ctls_mutex);
611 list_for_each(item, &mc_devices) {
612 mci = list_entry(item, struct mem_ctl_info, link);
614 if (mci->op_state == OP_RUNNING_POLL)
615 edac_mod_work(&mci->work, value);
617 mutex_unlock(&mem_ctls_mutex);
622 /* Return 0 on success, 1 on failure.
623 * Before calling this function, caller must
624 * assign a unique value to mci->mc_idx.
626 * locking model:
628 * called with the mem_ctls_mutex lock held
630 static int add_mc_to_global_list(struct mem_ctl_info *mci)
632 struct list_head *item, *insert_before;
633 struct mem_ctl_info *p;
635 insert_before = &mc_devices;
637 p = __find_mci_by_dev(mci->pdev);
638 if (unlikely(p != NULL))
639 goto fail0;
641 list_for_each(item, &mc_devices) {
642 p = list_entry(item, struct mem_ctl_info, link);
644 if (p->mc_idx >= mci->mc_idx) {
645 if (unlikely(p->mc_idx == mci->mc_idx))
646 goto fail1;
648 insert_before = item;
649 break;
653 list_add_tail_rcu(&mci->link, insert_before);
654 return 0;
656 fail0:
657 edac_printk(KERN_WARNING, EDAC_MC,
658 "%s (%s) %s %s already assigned %d\n", dev_name(p->pdev),
659 edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
660 return 1;
662 fail1:
663 edac_printk(KERN_WARNING, EDAC_MC,
664 "bug in low-level driver: attempt to assign\n"
665 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
666 return 1;
669 static int del_mc_from_global_list(struct mem_ctl_info *mci)
671 list_del_rcu(&mci->link);
673 /* these are for safe removal of devices from global list while
674 * NMI handlers may be traversing list
676 synchronize_rcu();
677 INIT_LIST_HEAD(&mci->link);
679 return list_empty(&mc_devices);
682 struct mem_ctl_info *edac_mc_find(int idx)
684 struct mem_ctl_info *mci;
685 struct list_head *item;
687 mutex_lock(&mem_ctls_mutex);
689 list_for_each(item, &mc_devices) {
690 mci = list_entry(item, struct mem_ctl_info, link);
691 if (mci->mc_idx == idx)
692 goto unlock;
695 mci = NULL;
696 unlock:
697 mutex_unlock(&mem_ctls_mutex);
698 return mci;
700 EXPORT_SYMBOL(edac_mc_find);
702 const char *edac_get_owner(void)
704 return edac_mc_owner;
706 EXPORT_SYMBOL_GPL(edac_get_owner);
708 /* FIXME - should a warning be printed if no error detection? correction? */
709 int edac_mc_add_mc_with_groups(struct mem_ctl_info *mci,
710 const struct attribute_group **groups)
712 int ret = -EINVAL;
713 edac_dbg(0, "\n");
715 if (mci->mc_idx >= EDAC_MAX_MCS) {
716 pr_warn_once("Too many memory controllers: %d\n", mci->mc_idx);
717 return -ENODEV;
720 #ifdef CONFIG_EDAC_DEBUG
721 if (edac_debug_level >= 3)
722 edac_mc_dump_mci(mci);
724 if (edac_debug_level >= 4) {
725 int i;
727 for (i = 0; i < mci->nr_csrows; i++) {
728 struct csrow_info *csrow = mci->csrows[i];
729 u32 nr_pages = 0;
730 int j;
732 for (j = 0; j < csrow->nr_channels; j++)
733 nr_pages += csrow->channels[j]->dimm->nr_pages;
734 if (!nr_pages)
735 continue;
736 edac_mc_dump_csrow(csrow);
737 for (j = 0; j < csrow->nr_channels; j++)
738 if (csrow->channels[j]->dimm->nr_pages)
739 edac_mc_dump_channel(csrow->channels[j]);
741 for (i = 0; i < mci->tot_dimms; i++)
742 if (mci->dimms[i]->nr_pages)
743 edac_mc_dump_dimm(mci->dimms[i], i);
745 #endif
746 mutex_lock(&mem_ctls_mutex);
748 if (edac_mc_owner && edac_mc_owner != mci->mod_name) {
749 ret = -EPERM;
750 goto fail0;
753 if (add_mc_to_global_list(mci))
754 goto fail0;
756 /* set load time so that error rate can be tracked */
757 mci->start_time = jiffies;
759 mci->bus = &mc_bus[mci->mc_idx];
761 if (edac_create_sysfs_mci_device(mci, groups)) {
762 edac_mc_printk(mci, KERN_WARNING,
763 "failed to create sysfs device\n");
764 goto fail1;
767 if (mci->edac_check) {
768 mci->op_state = OP_RUNNING_POLL;
770 INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
771 edac_queue_work(&mci->work, msecs_to_jiffies(edac_mc_get_poll_msec()));
773 } else {
774 mci->op_state = OP_RUNNING_INTERRUPT;
777 /* Report action taken */
778 edac_mc_printk(mci, KERN_INFO,
779 "Giving out device to module %s controller %s: DEV %s (%s)\n",
780 mci->mod_name, mci->ctl_name, mci->dev_name,
781 edac_op_state_to_string(mci->op_state));
783 edac_mc_owner = mci->mod_name;
785 mutex_unlock(&mem_ctls_mutex);
786 return 0;
788 fail1:
789 del_mc_from_global_list(mci);
791 fail0:
792 mutex_unlock(&mem_ctls_mutex);
793 return ret;
795 EXPORT_SYMBOL_GPL(edac_mc_add_mc_with_groups);
797 struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
799 struct mem_ctl_info *mci;
801 edac_dbg(0, "\n");
803 mutex_lock(&mem_ctls_mutex);
805 /* find the requested mci struct in the global list */
806 mci = __find_mci_by_dev(dev);
807 if (mci == NULL) {
808 mutex_unlock(&mem_ctls_mutex);
809 return NULL;
812 /* mark MCI offline: */
813 mci->op_state = OP_OFFLINE;
815 if (del_mc_from_global_list(mci))
816 edac_mc_owner = NULL;
818 mutex_unlock(&mem_ctls_mutex);
820 if (mci->edac_check)
821 edac_stop_work(&mci->work);
823 /* remove from sysfs */
824 edac_remove_sysfs_mci_device(mci);
826 edac_printk(KERN_INFO, EDAC_MC,
827 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
828 mci->mod_name, mci->ctl_name, edac_dev_name(mci));
830 return mci;
832 EXPORT_SYMBOL_GPL(edac_mc_del_mc);
834 static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
835 u32 size)
837 struct page *pg;
838 void *virt_addr;
839 unsigned long flags = 0;
841 edac_dbg(3, "\n");
843 /* ECC error page was not in our memory. Ignore it. */
844 if (!pfn_valid(page))
845 return;
847 /* Find the actual page structure then map it and fix */
848 pg = pfn_to_page(page);
850 if (PageHighMem(pg))
851 local_irq_save(flags);
853 virt_addr = kmap_atomic(pg);
855 /* Perform architecture specific atomic scrub operation */
856 edac_atomic_scrub(virt_addr + offset, size);
858 /* Unmap and complete */
859 kunmap_atomic(virt_addr);
861 if (PageHighMem(pg))
862 local_irq_restore(flags);
865 /* FIXME - should return -1 */
866 int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
868 struct csrow_info **csrows = mci->csrows;
869 int row, i, j, n;
871 edac_dbg(1, "MC%d: 0x%lx\n", mci->mc_idx, page);
872 row = -1;
874 for (i = 0; i < mci->nr_csrows; i++) {
875 struct csrow_info *csrow = csrows[i];
876 n = 0;
877 for (j = 0; j < csrow->nr_channels; j++) {
878 struct dimm_info *dimm = csrow->channels[j]->dimm;
879 n += dimm->nr_pages;
881 if (n == 0)
882 continue;
884 edac_dbg(3, "MC%d: first(0x%lx) page(0x%lx) last(0x%lx) mask(0x%lx)\n",
885 mci->mc_idx,
886 csrow->first_page, page, csrow->last_page,
887 csrow->page_mask);
889 if ((page >= csrow->first_page) &&
890 (page <= csrow->last_page) &&
891 ((page & csrow->page_mask) ==
892 (csrow->first_page & csrow->page_mask))) {
893 row = i;
894 break;
898 if (row == -1)
899 edac_mc_printk(mci, KERN_ERR,
900 "could not look up page error address %lx\n",
901 (unsigned long)page);
903 return row;
905 EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
907 const char *edac_layer_name[] = {
908 [EDAC_MC_LAYER_BRANCH] = "branch",
909 [EDAC_MC_LAYER_CHANNEL] = "channel",
910 [EDAC_MC_LAYER_SLOT] = "slot",
911 [EDAC_MC_LAYER_CHIP_SELECT] = "csrow",
912 [EDAC_MC_LAYER_ALL_MEM] = "memory",
914 EXPORT_SYMBOL_GPL(edac_layer_name);
916 static void edac_inc_ce_error(struct mem_ctl_info *mci,
917 bool enable_per_layer_report,
918 const int pos[EDAC_MAX_LAYERS],
919 const u16 count)
921 int i, index = 0;
923 mci->ce_mc += count;
925 if (!enable_per_layer_report) {
926 mci->ce_noinfo_count += count;
927 return;
930 for (i = 0; i < mci->n_layers; i++) {
931 if (pos[i] < 0)
932 break;
933 index += pos[i];
934 mci->ce_per_layer[i][index] += count;
936 if (i < mci->n_layers - 1)
937 index *= mci->layers[i + 1].size;
941 static void edac_inc_ue_error(struct mem_ctl_info *mci,
942 bool enable_per_layer_report,
943 const int pos[EDAC_MAX_LAYERS],
944 const u16 count)
946 int i, index = 0;
948 mci->ue_mc += count;
950 if (!enable_per_layer_report) {
951 mci->ue_noinfo_count += count;
952 return;
955 for (i = 0; i < mci->n_layers; i++) {
956 if (pos[i] < 0)
957 break;
958 index += pos[i];
959 mci->ue_per_layer[i][index] += count;
961 if (i < mci->n_layers - 1)
962 index *= mci->layers[i + 1].size;
966 static void edac_ce_error(struct mem_ctl_info *mci,
967 const u16 error_count,
968 const int pos[EDAC_MAX_LAYERS],
969 const char *msg,
970 const char *location,
971 const char *label,
972 const char *detail,
973 const char *other_detail,
974 const bool enable_per_layer_report,
975 const unsigned long page_frame_number,
976 const unsigned long offset_in_page,
977 long grain)
979 unsigned long remapped_page;
980 char *msg_aux = "";
982 if (*msg)
983 msg_aux = " ";
985 if (edac_mc_get_log_ce()) {
986 if (other_detail && *other_detail)
987 edac_mc_printk(mci, KERN_WARNING,
988 "%d CE %s%son %s (%s %s - %s)\n",
989 error_count, msg, msg_aux, label,
990 location, detail, other_detail);
991 else
992 edac_mc_printk(mci, KERN_WARNING,
993 "%d CE %s%son %s (%s %s)\n",
994 error_count, msg, msg_aux, label,
995 location, detail);
997 edac_inc_ce_error(mci, enable_per_layer_report, pos, error_count);
999 if (mci->scrub_mode == SCRUB_SW_SRC) {
1001 * Some memory controllers (called MCs below) can remap
1002 * memory so that it is still available at a different
1003 * address when PCI devices map into memory.
1004 * MC's that can't do this, lose the memory where PCI
1005 * devices are mapped. This mapping is MC-dependent
1006 * and so we call back into the MC driver for it to
1007 * map the MC page to a physical (CPU) page which can
1008 * then be mapped to a virtual page - which can then
1009 * be scrubbed.
1011 remapped_page = mci->ctl_page_to_phys ?
1012 mci->ctl_page_to_phys(mci, page_frame_number) :
1013 page_frame_number;
1015 edac_mc_scrub_block(remapped_page,
1016 offset_in_page, grain);
1020 static void edac_ue_error(struct mem_ctl_info *mci,
1021 const u16 error_count,
1022 const int pos[EDAC_MAX_LAYERS],
1023 const char *msg,
1024 const char *location,
1025 const char *label,
1026 const char *detail,
1027 const char *other_detail,
1028 const bool enable_per_layer_report)
1030 char *msg_aux = "";
1032 if (*msg)
1033 msg_aux = " ";
1035 if (edac_mc_get_log_ue()) {
1036 if (other_detail && *other_detail)
1037 edac_mc_printk(mci, KERN_WARNING,
1038 "%d UE %s%son %s (%s %s - %s)\n",
1039 error_count, msg, msg_aux, label,
1040 location, detail, other_detail);
1041 else
1042 edac_mc_printk(mci, KERN_WARNING,
1043 "%d UE %s%son %s (%s %s)\n",
1044 error_count, msg, msg_aux, label,
1045 location, detail);
1048 if (edac_mc_get_panic_on_ue()) {
1049 if (other_detail && *other_detail)
1050 panic("UE %s%son %s (%s%s - %s)\n",
1051 msg, msg_aux, label, location, detail, other_detail);
1052 else
1053 panic("UE %s%son %s (%s%s)\n",
1054 msg, msg_aux, label, location, detail);
1057 edac_inc_ue_error(mci, enable_per_layer_report, pos, error_count);
1060 void edac_raw_mc_handle_error(const enum hw_event_mc_err_type type,
1061 struct mem_ctl_info *mci,
1062 struct edac_raw_error_desc *e)
1064 char detail[80];
1065 int pos[EDAC_MAX_LAYERS] = { e->top_layer, e->mid_layer, e->low_layer };
1067 /* Memory type dependent details about the error */
1068 if (type == HW_EVENT_ERR_CORRECTED) {
1069 snprintf(detail, sizeof(detail),
1070 "page:0x%lx offset:0x%lx grain:%ld syndrome:0x%lx",
1071 e->page_frame_number, e->offset_in_page,
1072 e->grain, e->syndrome);
1073 edac_ce_error(mci, e->error_count, pos, e->msg, e->location, e->label,
1074 detail, e->other_detail, e->enable_per_layer_report,
1075 e->page_frame_number, e->offset_in_page, e->grain);
1076 } else {
1077 snprintf(detail, sizeof(detail),
1078 "page:0x%lx offset:0x%lx grain:%ld",
1079 e->page_frame_number, e->offset_in_page, e->grain);
1081 edac_ue_error(mci, e->error_count, pos, e->msg, e->location, e->label,
1082 detail, e->other_detail, e->enable_per_layer_report);
1087 EXPORT_SYMBOL_GPL(edac_raw_mc_handle_error);
1089 void edac_mc_handle_error(const enum hw_event_mc_err_type type,
1090 struct mem_ctl_info *mci,
1091 const u16 error_count,
1092 const unsigned long page_frame_number,
1093 const unsigned long offset_in_page,
1094 const unsigned long syndrome,
1095 const int top_layer,
1096 const int mid_layer,
1097 const int low_layer,
1098 const char *msg,
1099 const char *other_detail)
1101 char *p;
1102 int row = -1, chan = -1;
1103 int pos[EDAC_MAX_LAYERS] = { top_layer, mid_layer, low_layer };
1104 int i, n_labels = 0;
1105 u8 grain_bits;
1106 struct edac_raw_error_desc *e = &mci->error_desc;
1108 edac_dbg(3, "MC%d\n", mci->mc_idx);
1110 /* Fills the error report buffer */
1111 memset(e, 0, sizeof (*e));
1112 e->error_count = error_count;
1113 e->top_layer = top_layer;
1114 e->mid_layer = mid_layer;
1115 e->low_layer = low_layer;
1116 e->page_frame_number = page_frame_number;
1117 e->offset_in_page = offset_in_page;
1118 e->syndrome = syndrome;
1119 e->msg = msg;
1120 e->other_detail = other_detail;
1123 * Check if the event report is consistent and if the memory
1124 * location is known. If it is known, enable_per_layer_report will be
1125 * true, the DIMM(s) label info will be filled and the per-layer
1126 * error counters will be incremented.
1128 for (i = 0; i < mci->n_layers; i++) {
1129 if (pos[i] >= (int)mci->layers[i].size) {
1131 edac_mc_printk(mci, KERN_ERR,
1132 "INTERNAL ERROR: %s value is out of range (%d >= %d)\n",
1133 edac_layer_name[mci->layers[i].type],
1134 pos[i], mci->layers[i].size);
1136 * Instead of just returning it, let's use what's
1137 * known about the error. The increment routines and
1138 * the DIMM filter logic will do the right thing by
1139 * pointing the likely damaged DIMMs.
1141 pos[i] = -1;
1143 if (pos[i] >= 0)
1144 e->enable_per_layer_report = true;
1148 * Get the dimm label/grain that applies to the match criteria.
1149 * As the error algorithm may not be able to point to just one memory
1150 * stick, the logic here will get all possible labels that could
1151 * pottentially be affected by the error.
1152 * On FB-DIMM memory controllers, for uncorrected errors, it is common
1153 * to have only the MC channel and the MC dimm (also called "branch")
1154 * but the channel is not known, as the memory is arranged in pairs,
1155 * where each memory belongs to a separate channel within the same
1156 * branch.
1158 p = e->label;
1159 *p = '\0';
1161 for (i = 0; i < mci->tot_dimms; i++) {
1162 struct dimm_info *dimm = mci->dimms[i];
1164 if (top_layer >= 0 && top_layer != dimm->location[0])
1165 continue;
1166 if (mid_layer >= 0 && mid_layer != dimm->location[1])
1167 continue;
1168 if (low_layer >= 0 && low_layer != dimm->location[2])
1169 continue;
1171 /* get the max grain, over the error match range */
1172 if (dimm->grain > e->grain)
1173 e->grain = dimm->grain;
1176 * If the error is memory-controller wide, there's no need to
1177 * seek for the affected DIMMs because the whole
1178 * channel/memory controller/... may be affected.
1179 * Also, don't show errors for empty DIMM slots.
1181 if (e->enable_per_layer_report && dimm->nr_pages) {
1182 if (n_labels >= EDAC_MAX_LABELS) {
1183 e->enable_per_layer_report = false;
1184 break;
1186 n_labels++;
1187 if (p != e->label) {
1188 strcpy(p, OTHER_LABEL);
1189 p += strlen(OTHER_LABEL);
1191 strcpy(p, dimm->label);
1192 p += strlen(p);
1193 *p = '\0';
1196 * get csrow/channel of the DIMM, in order to allow
1197 * incrementing the compat API counters
1199 edac_dbg(4, "%s csrows map: (%d,%d)\n",
1200 mci->csbased ? "rank" : "dimm",
1201 dimm->csrow, dimm->cschannel);
1202 if (row == -1)
1203 row = dimm->csrow;
1204 else if (row >= 0 && row != dimm->csrow)
1205 row = -2;
1207 if (chan == -1)
1208 chan = dimm->cschannel;
1209 else if (chan >= 0 && chan != dimm->cschannel)
1210 chan = -2;
1214 if (!e->enable_per_layer_report) {
1215 strcpy(e->label, "any memory");
1216 } else {
1217 edac_dbg(4, "csrow/channel to increment: (%d,%d)\n", row, chan);
1218 if (p == e->label)
1219 strcpy(e->label, "unknown memory");
1220 if (type == HW_EVENT_ERR_CORRECTED) {
1221 if (row >= 0) {
1222 mci->csrows[row]->ce_count += error_count;
1223 if (chan >= 0)
1224 mci->csrows[row]->channels[chan]->ce_count += error_count;
1226 } else
1227 if (row >= 0)
1228 mci->csrows[row]->ue_count += error_count;
1231 /* Fill the RAM location data */
1232 p = e->location;
1234 for (i = 0; i < mci->n_layers; i++) {
1235 if (pos[i] < 0)
1236 continue;
1238 p += sprintf(p, "%s:%d ",
1239 edac_layer_name[mci->layers[i].type],
1240 pos[i]);
1242 if (p > e->location)
1243 *(p - 1) = '\0';
1245 /* Sanity-check driver-supplied grain value. */
1246 if (WARN_ON_ONCE(!e->grain))
1247 e->grain = 1;
1249 grain_bits = fls_long(e->grain - 1);
1251 /* Report the error via the trace interface */
1252 if (IS_ENABLED(CONFIG_RAS))
1253 trace_mc_event(type, e->msg, e->label, e->error_count,
1254 mci->mc_idx, e->top_layer, e->mid_layer,
1255 e->low_layer,
1256 (e->page_frame_number << PAGE_SHIFT) | e->offset_in_page,
1257 grain_bits, e->syndrome, e->other_detail);
1259 edac_raw_mc_handle_error(type, mci, e);
1261 EXPORT_SYMBOL_GPL(edac_mc_handle_error);