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>
34 #include "edac_module.h"
35 #include <ras/ras_event.h>
37 #ifdef CONFIG_EDAC_ATOMIC_SCRUB
40 #define edac_atomic_scrub(va, size) do { } while (0)
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 void const *edac_mc_owner;
58 static struct bus_type mc_bus[EDAC_MAX_MCS];
60 int edac_get_report_status(void)
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)
73 EXPORT_SYMBOL_GPL(edac_set_report_status);
75 static int edac_report_set(const char *str, const struct kernel_param *kp)
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;
90 static int edac_report_get(char *buffer, const struct kernel_param *kp)
94 switch (edac_report) {
95 case EDAC_REPORTING_ENABLED:
96 ret = sprintf(buffer, "on");
98 case EDAC_REPORTING_DISABLED:
99 ret = sprintf(buffer, "off");
101 case EDAC_REPORTING_FORCE:
102 ret = sprintf(buffer, "force");
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,
122 struct mem_ctl_info *mci = dimm->mci;
126 for (i = 0; i < mci->n_layers; i++) {
127 n = snprintf(p, len, "%s %d ",
128 edac_layer_name[mci->layers[i].type],
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)
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 csrow",
199 [MEM_RESERVED] = "Reserved csrow type",
200 [MEM_UNKNOWN] = "Unknown csrow type",
201 [MEM_FPM] = "Fast page mode RAM",
202 [MEM_EDO] = "Extended data out RAM",
203 [MEM_BEDO] = "Burst Extended data out RAM",
204 [MEM_SDR] = "Single data rate SDRAM",
205 [MEM_RDR] = "Registered single data rate SDRAM",
206 [MEM_DDR] = "Double data rate SDRAM",
207 [MEM_RDDR] = "Registered Double data rate SDRAM",
208 [MEM_RMBS] = "Rambus DRAM",
209 [MEM_DDR2] = "Unbuffered DDR2 RAM",
210 [MEM_FB_DDR2] = "Fully buffered DDR2",
211 [MEM_RDDR2] = "Registered DDR2 RAM",
212 [MEM_XDR] = "Rambus XDR",
213 [MEM_DDR3] = "Unbuffered DDR3 RAM",
214 [MEM_RDDR3] = "Registered DDR3 RAM",
215 [MEM_LRDDR3] = "Load-Reduced DDR3 RAM",
216 [MEM_DDR4] = "Unbuffered DDR4 RAM",
217 [MEM_RDDR4] = "Registered DDR4 RAM",
218 [MEM_LRDDR4] = "Load-Reduced-DDR4-RAM",
220 EXPORT_SYMBOL_GPL(edac_mem_types);
223 * edac_align_ptr - Prepares the pointer offsets for a single-shot allocation
224 * @p: pointer to a pointer with the memory offset to be used. At
225 * return, this will be incremented to point to the next offset
226 * @size: Size of the data structure to be reserved
227 * @n_elems: Number of elements that should be reserved
229 * If 'size' is a constant, the compiler will optimize this whole function
230 * down to either a no-op or the addition of a constant to the value of '*p'.
232 * The 'p' pointer is absolutely needed to keep the proper advancing
233 * further in memory to the proper offsets when allocating the struct along
234 * with its embedded structs, as edac_device_alloc_ctl_info() does it
235 * above, for example.
237 * At return, the pointer 'p' will be incremented to be used on a next call
240 void *edac_align_ptr(void **p, unsigned size, int n_elems)
245 *p += size * n_elems;
248 * 'p' can possibly be an unaligned item X such that sizeof(X) is
249 * 'size'. Adjust 'p' so that its alignment is at least as
250 * stringent as what the compiler would provide for X and return
251 * the aligned result.
252 * Here we assume that the alignment of a "long long" is the most
253 * stringent alignment that the compiler will ever provide by default.
254 * As far as I know, this is a reasonable assumption.
256 if (size > sizeof(long))
257 align = sizeof(long long);
258 else if (size > sizeof(int))
259 align = sizeof(long);
260 else if (size > sizeof(short))
262 else if (size > sizeof(char))
263 align = sizeof(short);
267 r = (unsigned long)ptr % align;
274 return (void *)(((unsigned long)ptr) + align - r);
277 static void _edac_mc_free(struct mem_ctl_info *mci)
280 struct csrow_info *csr;
281 const unsigned int tot_dimms = mci->tot_dimms;
282 const unsigned int tot_channels = mci->num_cschannel;
283 const unsigned int tot_csrows = mci->nr_csrows;
286 for (i = 0; i < tot_dimms; i++)
287 kfree(mci->dimms[i]);
291 for (row = 0; row < tot_csrows; row++) {
292 csr = mci->csrows[row];
295 for (chn = 0; chn < tot_channels; chn++)
296 kfree(csr->channels[chn]);
297 kfree(csr->channels);
307 struct mem_ctl_info *edac_mc_alloc(unsigned mc_num,
309 struct edac_mc_layer *layers,
312 struct mem_ctl_info *mci;
313 struct edac_mc_layer *layer;
314 struct csrow_info *csr;
315 struct rank_info *chan;
316 struct dimm_info *dimm;
317 u32 *ce_per_layer[EDAC_MAX_LAYERS], *ue_per_layer[EDAC_MAX_LAYERS];
318 unsigned pos[EDAC_MAX_LAYERS];
319 unsigned size, tot_dimms = 1, count = 1;
320 unsigned tot_csrows = 1, tot_channels = 1, tot_errcount = 0;
321 void *pvt, *p, *ptr = NULL;
322 int i, j, row, chn, n, len, off;
323 bool per_rank = false;
325 BUG_ON(n_layers > EDAC_MAX_LAYERS || n_layers == 0);
327 * Calculate the total amount of dimms and csrows/cschannels while
328 * in the old API emulation mode
330 for (i = 0; i < n_layers; i++) {
331 tot_dimms *= layers[i].size;
332 if (layers[i].is_virt_csrow)
333 tot_csrows *= layers[i].size;
335 tot_channels *= layers[i].size;
337 if (layers[i].type == EDAC_MC_LAYER_CHIP_SELECT)
341 /* Figure out the offsets of the various items from the start of an mc
342 * structure. We want the alignment of each item to be at least as
343 * stringent as what the compiler would provide if we could simply
344 * hardcode everything into a single struct.
346 mci = edac_align_ptr(&ptr, sizeof(*mci), 1);
347 layer = edac_align_ptr(&ptr, sizeof(*layer), n_layers);
348 for (i = 0; i < n_layers; i++) {
349 count *= layers[i].size;
350 edac_dbg(4, "errcount layer %d size %d\n", i, count);
351 ce_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count);
352 ue_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count);
353 tot_errcount += 2 * count;
356 edac_dbg(4, "allocating %d error counters\n", tot_errcount);
357 pvt = edac_align_ptr(&ptr, sz_pvt, 1);
358 size = ((unsigned long)pvt) + sz_pvt;
360 edac_dbg(1, "allocating %u bytes for mci data (%d %s, %d csrows/channels)\n",
363 per_rank ? "ranks" : "dimms",
364 tot_csrows * tot_channels);
366 mci = kzalloc(size, GFP_KERNEL);
370 /* Adjust pointers so they point within the memory we just allocated
371 * rather than an imaginary chunk of memory located at address 0.
373 layer = (struct edac_mc_layer *)(((char *)mci) + ((unsigned long)layer));
374 for (i = 0; i < n_layers; i++) {
375 mci->ce_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ce_per_layer[i]));
376 mci->ue_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ue_per_layer[i]));
378 pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL;
380 /* setup index and various internal pointers */
381 mci->mc_idx = mc_num;
382 mci->tot_dimms = tot_dimms;
384 mci->n_layers = n_layers;
386 memcpy(mci->layers, layers, sizeof(*layer) * n_layers);
387 mci->nr_csrows = tot_csrows;
388 mci->num_cschannel = tot_channels;
389 mci->csbased = per_rank;
392 * Alocate and fill the csrow/channels structs
394 mci->csrows = kcalloc(tot_csrows, sizeof(*mci->csrows), GFP_KERNEL);
397 for (row = 0; row < tot_csrows; row++) {
398 csr = kzalloc(sizeof(**mci->csrows), GFP_KERNEL);
401 mci->csrows[row] = csr;
402 csr->csrow_idx = row;
404 csr->nr_channels = tot_channels;
405 csr->channels = kcalloc(tot_channels, sizeof(*csr->channels),
410 for (chn = 0; chn < tot_channels; chn++) {
411 chan = kzalloc(sizeof(**csr->channels), GFP_KERNEL);
414 csr->channels[chn] = chan;
415 chan->chan_idx = chn;
421 * Allocate and fill the dimm structs
423 mci->dimms = kcalloc(tot_dimms, sizeof(*mci->dimms), GFP_KERNEL);
427 memset(&pos, 0, sizeof(pos));
430 for (i = 0; i < tot_dimms; i++) {
431 chan = mci->csrows[row]->channels[chn];
432 off = EDAC_DIMM_OFF(layer, n_layers, pos[0], pos[1], pos[2]);
433 if (off < 0 || off >= tot_dimms) {
434 edac_mc_printk(mci, KERN_ERR, "EDAC core bug: EDAC_DIMM_OFF is trying to do an illegal data access\n");
438 dimm = kzalloc(sizeof(**mci->dimms), GFP_KERNEL);
441 mci->dimms[off] = dimm;
445 * Copy DIMM location and initialize it.
447 len = sizeof(dimm->label);
449 n = snprintf(p, len, "mc#%u", mc_num);
452 for (j = 0; j < n_layers; j++) {
453 n = snprintf(p, len, "%s#%u",
454 edac_layer_name[layers[j].type],
458 dimm->location[j] = pos[j];
464 /* Link it to the csrows old API data */
467 dimm->cschannel = chn;
469 /* Increment csrow location */
470 if (layers[0].is_virt_csrow) {
472 if (chn == tot_channels) {
478 if (row == tot_csrows) {
484 /* Increment dimm location */
485 for (j = n_layers - 1; j >= 0; j--) {
487 if (pos[j] < layers[j].size)
493 mci->op_state = OP_ALLOC;
502 EXPORT_SYMBOL_GPL(edac_mc_alloc);
504 void edac_mc_free(struct mem_ctl_info *mci)
508 /* If we're not yet registered with sysfs free only what was allocated
509 * in edac_mc_alloc().
511 if (!device_is_registered(&mci->dev)) {
516 /* the mci instance is freed here, when the sysfs object is dropped */
517 edac_unregister_sysfs(mci);
519 EXPORT_SYMBOL_GPL(edac_mc_free);
521 bool edac_has_mcs(void)
525 mutex_lock(&mem_ctls_mutex);
527 ret = list_empty(&mc_devices);
529 mutex_unlock(&mem_ctls_mutex);
533 EXPORT_SYMBOL_GPL(edac_has_mcs);
535 /* Caller must hold mem_ctls_mutex */
536 static struct mem_ctl_info *__find_mci_by_dev(struct device *dev)
538 struct mem_ctl_info *mci;
539 struct list_head *item;
543 list_for_each(item, &mc_devices) {
544 mci = list_entry(item, struct mem_ctl_info, link);
546 if (mci->pdev == dev)
556 * scan list of controllers looking for the one that manages
558 * @dev: pointer to a struct device related with the MCI
560 struct mem_ctl_info *find_mci_by_dev(struct device *dev)
562 struct mem_ctl_info *ret;
564 mutex_lock(&mem_ctls_mutex);
565 ret = __find_mci_by_dev(dev);
566 mutex_unlock(&mem_ctls_mutex);
570 EXPORT_SYMBOL_GPL(find_mci_by_dev);
573 * edac_mc_workq_function
574 * performs the operation scheduled by a workq request
576 static void edac_mc_workq_function(struct work_struct *work_req)
578 struct delayed_work *d_work = to_delayed_work(work_req);
579 struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
581 mutex_lock(&mem_ctls_mutex);
583 if (mci->op_state != OP_RUNNING_POLL) {
584 mutex_unlock(&mem_ctls_mutex);
588 if (edac_op_state == EDAC_OPSTATE_POLL)
589 mci->edac_check(mci);
591 mutex_unlock(&mem_ctls_mutex);
593 /* Queue ourselves again. */
594 edac_queue_work(&mci->work, msecs_to_jiffies(edac_mc_get_poll_msec()));
598 * edac_mc_reset_delay_period(unsigned long value)
600 * user space has updated our poll period value, need to
601 * reset our workq delays
603 void edac_mc_reset_delay_period(unsigned long value)
605 struct mem_ctl_info *mci;
606 struct list_head *item;
608 mutex_lock(&mem_ctls_mutex);
610 list_for_each(item, &mc_devices) {
611 mci = list_entry(item, struct mem_ctl_info, link);
613 if (mci->op_state == OP_RUNNING_POLL)
614 edac_mod_work(&mci->work, value);
616 mutex_unlock(&mem_ctls_mutex);
621 /* Return 0 on success, 1 on failure.
622 * Before calling this function, caller must
623 * assign a unique value to mci->mc_idx.
627 * called with the mem_ctls_mutex lock held
629 static int add_mc_to_global_list(struct mem_ctl_info *mci)
631 struct list_head *item, *insert_before;
632 struct mem_ctl_info *p;
634 insert_before = &mc_devices;
636 p = __find_mci_by_dev(mci->pdev);
637 if (unlikely(p != NULL))
640 list_for_each(item, &mc_devices) {
641 p = list_entry(item, struct mem_ctl_info, link);
643 if (p->mc_idx >= mci->mc_idx) {
644 if (unlikely(p->mc_idx == mci->mc_idx))
647 insert_before = item;
652 list_add_tail_rcu(&mci->link, insert_before);
656 edac_printk(KERN_WARNING, EDAC_MC,
657 "%s (%s) %s %s already assigned %d\n", dev_name(p->pdev),
658 edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
662 edac_printk(KERN_WARNING, EDAC_MC,
663 "bug in low-level driver: attempt to assign\n"
664 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
668 static int del_mc_from_global_list(struct mem_ctl_info *mci)
670 list_del_rcu(&mci->link);
672 /* these are for safe removal of devices from global list while
673 * NMI handlers may be traversing list
676 INIT_LIST_HEAD(&mci->link);
678 return list_empty(&mc_devices);
681 struct mem_ctl_info *edac_mc_find(int idx)
683 struct mem_ctl_info *mci;
684 struct list_head *item;
686 mutex_lock(&mem_ctls_mutex);
688 list_for_each(item, &mc_devices) {
689 mci = list_entry(item, struct mem_ctl_info, link);
690 if (mci->mc_idx == idx)
696 mutex_unlock(&mem_ctls_mutex);
699 EXPORT_SYMBOL(edac_mc_find);
702 /* FIXME - should a warning be printed if no error detection? correction? */
703 int edac_mc_add_mc_with_groups(struct mem_ctl_info *mci,
704 const struct attribute_group **groups)
709 if (mci->mc_idx >= EDAC_MAX_MCS) {
710 pr_warn_once("Too many memory controllers: %d\n", mci->mc_idx);
714 #ifdef CONFIG_EDAC_DEBUG
715 if (edac_debug_level >= 3)
716 edac_mc_dump_mci(mci);
718 if (edac_debug_level >= 4) {
721 for (i = 0; i < mci->nr_csrows; i++) {
722 struct csrow_info *csrow = mci->csrows[i];
726 for (j = 0; j < csrow->nr_channels; j++)
727 nr_pages += csrow->channels[j]->dimm->nr_pages;
730 edac_mc_dump_csrow(csrow);
731 for (j = 0; j < csrow->nr_channels; j++)
732 if (csrow->channels[j]->dimm->nr_pages)
733 edac_mc_dump_channel(csrow->channels[j]);
735 for (i = 0; i < mci->tot_dimms; i++)
736 if (mci->dimms[i]->nr_pages)
737 edac_mc_dump_dimm(mci->dimms[i], i);
740 mutex_lock(&mem_ctls_mutex);
742 if (edac_mc_owner && edac_mc_owner != mci->mod_name) {
747 if (add_mc_to_global_list(mci))
750 /* set load time so that error rate can be tracked */
751 mci->start_time = jiffies;
753 mci->bus = &mc_bus[mci->mc_idx];
755 if (edac_create_sysfs_mci_device(mci, groups)) {
756 edac_mc_printk(mci, KERN_WARNING,
757 "failed to create sysfs device\n");
761 if (mci->edac_check) {
762 mci->op_state = OP_RUNNING_POLL;
764 INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
765 edac_queue_work(&mci->work, msecs_to_jiffies(edac_mc_get_poll_msec()));
768 mci->op_state = OP_RUNNING_INTERRUPT;
771 /* Report action taken */
772 edac_mc_printk(mci, KERN_INFO,
773 "Giving out device to module %s controller %s: DEV %s (%s)\n",
774 mci->mod_name, mci->ctl_name, mci->dev_name,
775 edac_op_state_to_string(mci->op_state));
777 edac_mc_owner = mci->mod_name;
779 mutex_unlock(&mem_ctls_mutex);
783 del_mc_from_global_list(mci);
786 mutex_unlock(&mem_ctls_mutex);
789 EXPORT_SYMBOL_GPL(edac_mc_add_mc_with_groups);
791 struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
793 struct mem_ctl_info *mci;
797 mutex_lock(&mem_ctls_mutex);
799 /* find the requested mci struct in the global list */
800 mci = __find_mci_by_dev(dev);
802 mutex_unlock(&mem_ctls_mutex);
806 /* mark MCI offline: */
807 mci->op_state = OP_OFFLINE;
809 if (del_mc_from_global_list(mci))
810 edac_mc_owner = NULL;
812 mutex_unlock(&mem_ctls_mutex);
815 edac_stop_work(&mci->work);
817 /* remove from sysfs */
818 edac_remove_sysfs_mci_device(mci);
820 edac_printk(KERN_INFO, EDAC_MC,
821 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
822 mci->mod_name, mci->ctl_name, edac_dev_name(mci));
826 EXPORT_SYMBOL_GPL(edac_mc_del_mc);
828 static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
833 unsigned long flags = 0;
837 /* ECC error page was not in our memory. Ignore it. */
838 if (!pfn_valid(page))
841 /* Find the actual page structure then map it and fix */
842 pg = pfn_to_page(page);
845 local_irq_save(flags);
847 virt_addr = kmap_atomic(pg);
849 /* Perform architecture specific atomic scrub operation */
850 edac_atomic_scrub(virt_addr + offset, size);
852 /* Unmap and complete */
853 kunmap_atomic(virt_addr);
856 local_irq_restore(flags);
859 /* FIXME - should return -1 */
860 int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
862 struct csrow_info **csrows = mci->csrows;
865 edac_dbg(1, "MC%d: 0x%lx\n", mci->mc_idx, page);
868 for (i = 0; i < mci->nr_csrows; i++) {
869 struct csrow_info *csrow = csrows[i];
871 for (j = 0; j < csrow->nr_channels; j++) {
872 struct dimm_info *dimm = csrow->channels[j]->dimm;
878 edac_dbg(3, "MC%d: first(0x%lx) page(0x%lx) last(0x%lx) mask(0x%lx)\n",
880 csrow->first_page, page, csrow->last_page,
883 if ((page >= csrow->first_page) &&
884 (page <= csrow->last_page) &&
885 ((page & csrow->page_mask) ==
886 (csrow->first_page & csrow->page_mask))) {
893 edac_mc_printk(mci, KERN_ERR,
894 "could not look up page error address %lx\n",
895 (unsigned long)page);
899 EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
901 const char *edac_layer_name[] = {
902 [EDAC_MC_LAYER_BRANCH] = "branch",
903 [EDAC_MC_LAYER_CHANNEL] = "channel",
904 [EDAC_MC_LAYER_SLOT] = "slot",
905 [EDAC_MC_LAYER_CHIP_SELECT] = "csrow",
906 [EDAC_MC_LAYER_ALL_MEM] = "memory",
908 EXPORT_SYMBOL_GPL(edac_layer_name);
910 static void edac_inc_ce_error(struct mem_ctl_info *mci,
911 bool enable_per_layer_report,
912 const int pos[EDAC_MAX_LAYERS],
919 if (!enable_per_layer_report) {
920 mci->ce_noinfo_count += count;
924 for (i = 0; i < mci->n_layers; i++) {
928 mci->ce_per_layer[i][index] += count;
930 if (i < mci->n_layers - 1)
931 index *= mci->layers[i + 1].size;
935 static void edac_inc_ue_error(struct mem_ctl_info *mci,
936 bool enable_per_layer_report,
937 const int pos[EDAC_MAX_LAYERS],
944 if (!enable_per_layer_report) {
945 mci->ue_noinfo_count += count;
949 for (i = 0; i < mci->n_layers; i++) {
953 mci->ue_per_layer[i][index] += count;
955 if (i < mci->n_layers - 1)
956 index *= mci->layers[i + 1].size;
960 static void edac_ce_error(struct mem_ctl_info *mci,
961 const u16 error_count,
962 const int pos[EDAC_MAX_LAYERS],
964 const char *location,
967 const char *other_detail,
968 const bool enable_per_layer_report,
969 const unsigned long page_frame_number,
970 const unsigned long offset_in_page,
973 unsigned long remapped_page;
979 if (edac_mc_get_log_ce()) {
980 if (other_detail && *other_detail)
981 edac_mc_printk(mci, KERN_WARNING,
982 "%d CE %s%son %s (%s %s - %s)\n",
983 error_count, msg, msg_aux, label,
984 location, detail, other_detail);
986 edac_mc_printk(mci, KERN_WARNING,
987 "%d CE %s%son %s (%s %s)\n",
988 error_count, msg, msg_aux, label,
991 edac_inc_ce_error(mci, enable_per_layer_report, pos, error_count);
993 if (mci->scrub_mode == SCRUB_SW_SRC) {
995 * Some memory controllers (called MCs below) can remap
996 * memory so that it is still available at a different
997 * address when PCI devices map into memory.
998 * MC's that can't do this, lose the memory where PCI
999 * devices are mapped. This mapping is MC-dependent
1000 * and so we call back into the MC driver for it to
1001 * map the MC page to a physical (CPU) page which can
1002 * then be mapped to a virtual page - which can then
1005 remapped_page = mci->ctl_page_to_phys ?
1006 mci->ctl_page_to_phys(mci, page_frame_number) :
1009 edac_mc_scrub_block(remapped_page,
1010 offset_in_page, grain);
1014 static void edac_ue_error(struct mem_ctl_info *mci,
1015 const u16 error_count,
1016 const int pos[EDAC_MAX_LAYERS],
1018 const char *location,
1021 const char *other_detail,
1022 const bool enable_per_layer_report)
1029 if (edac_mc_get_log_ue()) {
1030 if (other_detail && *other_detail)
1031 edac_mc_printk(mci, KERN_WARNING,
1032 "%d UE %s%son %s (%s %s - %s)\n",
1033 error_count, msg, msg_aux, label,
1034 location, detail, other_detail);
1036 edac_mc_printk(mci, KERN_WARNING,
1037 "%d UE %s%son %s (%s %s)\n",
1038 error_count, msg, msg_aux, label,
1042 if (edac_mc_get_panic_on_ue()) {
1043 if (other_detail && *other_detail)
1044 panic("UE %s%son %s (%s%s - %s)\n",
1045 msg, msg_aux, label, location, detail, other_detail);
1047 panic("UE %s%son %s (%s%s)\n",
1048 msg, msg_aux, label, location, detail);
1051 edac_inc_ue_error(mci, enable_per_layer_report, pos, error_count);
1054 void edac_raw_mc_handle_error(const enum hw_event_mc_err_type type,
1055 struct mem_ctl_info *mci,
1056 struct edac_raw_error_desc *e)
1059 int pos[EDAC_MAX_LAYERS] = { e->top_layer, e->mid_layer, e->low_layer };
1061 /* Memory type dependent details about the error */
1062 if (type == HW_EVENT_ERR_CORRECTED) {
1063 snprintf(detail, sizeof(detail),
1064 "page:0x%lx offset:0x%lx grain:%ld syndrome:0x%lx",
1065 e->page_frame_number, e->offset_in_page,
1066 e->grain, e->syndrome);
1067 edac_ce_error(mci, e->error_count, pos, e->msg, e->location, e->label,
1068 detail, e->other_detail, e->enable_per_layer_report,
1069 e->page_frame_number, e->offset_in_page, e->grain);
1071 snprintf(detail, sizeof(detail),
1072 "page:0x%lx offset:0x%lx grain:%ld",
1073 e->page_frame_number, e->offset_in_page, e->grain);
1075 edac_ue_error(mci, e->error_count, pos, e->msg, e->location, e->label,
1076 detail, e->other_detail, e->enable_per_layer_report);
1081 EXPORT_SYMBOL_GPL(edac_raw_mc_handle_error);
1083 void edac_mc_handle_error(const enum hw_event_mc_err_type type,
1084 struct mem_ctl_info *mci,
1085 const u16 error_count,
1086 const unsigned long page_frame_number,
1087 const unsigned long offset_in_page,
1088 const unsigned long syndrome,
1089 const int top_layer,
1090 const int mid_layer,
1091 const int low_layer,
1093 const char *other_detail)
1096 int row = -1, chan = -1;
1097 int pos[EDAC_MAX_LAYERS] = { top_layer, mid_layer, low_layer };
1098 int i, n_labels = 0;
1100 struct edac_raw_error_desc *e = &mci->error_desc;
1102 edac_dbg(3, "MC%d\n", mci->mc_idx);
1104 /* Fills the error report buffer */
1105 memset(e, 0, sizeof (*e));
1106 e->error_count = error_count;
1107 e->top_layer = top_layer;
1108 e->mid_layer = mid_layer;
1109 e->low_layer = low_layer;
1110 e->page_frame_number = page_frame_number;
1111 e->offset_in_page = offset_in_page;
1112 e->syndrome = syndrome;
1114 e->other_detail = other_detail;
1117 * Check if the event report is consistent and if the memory
1118 * location is known. If it is known, enable_per_layer_report will be
1119 * true, the DIMM(s) label info will be filled and the per-layer
1120 * error counters will be incremented.
1122 for (i = 0; i < mci->n_layers; i++) {
1123 if (pos[i] >= (int)mci->layers[i].size) {
1125 edac_mc_printk(mci, KERN_ERR,
1126 "INTERNAL ERROR: %s value is out of range (%d >= %d)\n",
1127 edac_layer_name[mci->layers[i].type],
1128 pos[i], mci->layers[i].size);
1130 * Instead of just returning it, let's use what's
1131 * known about the error. The increment routines and
1132 * the DIMM filter logic will do the right thing by
1133 * pointing the likely damaged DIMMs.
1138 e->enable_per_layer_report = true;
1142 * Get the dimm label/grain that applies to the match criteria.
1143 * As the error algorithm may not be able to point to just one memory
1144 * stick, the logic here will get all possible labels that could
1145 * pottentially be affected by the error.
1146 * On FB-DIMM memory controllers, for uncorrected errors, it is common
1147 * to have only the MC channel and the MC dimm (also called "branch")
1148 * but the channel is not known, as the memory is arranged in pairs,
1149 * where each memory belongs to a separate channel within the same
1155 for (i = 0; i < mci->tot_dimms; i++) {
1156 struct dimm_info *dimm = mci->dimms[i];
1158 if (top_layer >= 0 && top_layer != dimm->location[0])
1160 if (mid_layer >= 0 && mid_layer != dimm->location[1])
1162 if (low_layer >= 0 && low_layer != dimm->location[2])
1165 /* get the max grain, over the error match range */
1166 if (dimm->grain > e->grain)
1167 e->grain = dimm->grain;
1170 * If the error is memory-controller wide, there's no need to
1171 * seek for the affected DIMMs because the whole
1172 * channel/memory controller/... may be affected.
1173 * Also, don't show errors for empty DIMM slots.
1175 if (e->enable_per_layer_report && dimm->nr_pages) {
1176 if (n_labels >= EDAC_MAX_LABELS) {
1177 e->enable_per_layer_report = false;
1181 if (p != e->label) {
1182 strcpy(p, OTHER_LABEL);
1183 p += strlen(OTHER_LABEL);
1185 strcpy(p, dimm->label);
1190 * get csrow/channel of the DIMM, in order to allow
1191 * incrementing the compat API counters
1193 edac_dbg(4, "%s csrows map: (%d,%d)\n",
1194 mci->csbased ? "rank" : "dimm",
1195 dimm->csrow, dimm->cschannel);
1198 else if (row >= 0 && row != dimm->csrow)
1202 chan = dimm->cschannel;
1203 else if (chan >= 0 && chan != dimm->cschannel)
1208 if (!e->enable_per_layer_report) {
1209 strcpy(e->label, "any memory");
1211 edac_dbg(4, "csrow/channel to increment: (%d,%d)\n", row, chan);
1213 strcpy(e->label, "unknown memory");
1214 if (type == HW_EVENT_ERR_CORRECTED) {
1216 mci->csrows[row]->ce_count += error_count;
1218 mci->csrows[row]->channels[chan]->ce_count += error_count;
1222 mci->csrows[row]->ue_count += error_count;
1225 /* Fill the RAM location data */
1228 for (i = 0; i < mci->n_layers; i++) {
1232 p += sprintf(p, "%s:%d ",
1233 edac_layer_name[mci->layers[i].type],
1236 if (p > e->location)
1239 /* Sanity-check driver-supplied grain value. */
1240 if (WARN_ON_ONCE(!e->grain))
1243 grain_bits = fls_long(e->grain - 1);
1245 /* Report the error via the trace interface */
1246 if (IS_ENABLED(CONFIG_RAS))
1247 trace_mc_event(type, e->msg, e->label, e->error_count,
1248 mci->mc_idx, e->top_layer, e->mid_layer,
1250 (e->page_frame_number << PAGE_SHIFT) | e->offset_in_page,
1251 grain_bits, e->syndrome, e->other_detail);
1253 edac_raw_mc_handle_error(type, mci, e);
1255 EXPORT_SYMBOL_GPL(edac_mc_handle_error);