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 <asm/uaccess.h>
34 #include "edac_core.h"
35 #include "edac_module.h"
37 #define CREATE_TRACE_POINTS
38 #define TRACE_INCLUDE_PATH ../../include/ras
39 #include <ras/ras_event.h>
41 /* lock to memory controller's control array */
42 static DEFINE_MUTEX(mem_ctls_mutex);
43 static LIST_HEAD(mc_devices);
45 unsigned edac_dimm_info_location(struct dimm_info *dimm, char *buf,
48 struct mem_ctl_info *mci = dimm->mci;
52 for (i = 0; i < mci->n_layers; i++) {
53 n = snprintf(p, len, "%s %d ",
54 edac_layer_name[mci->layers[i].type],
66 #ifdef CONFIG_EDAC_DEBUG
68 static void edac_mc_dump_channel(struct rank_info *chan)
70 edac_dbg(4, " channel->chan_idx = %d\n", chan->chan_idx);
71 edac_dbg(4, " channel = %p\n", chan);
72 edac_dbg(4, " channel->csrow = %p\n", chan->csrow);
73 edac_dbg(4, " channel->dimm = %p\n", chan->dimm);
76 static void edac_mc_dump_dimm(struct dimm_info *dimm, int number)
80 edac_dimm_info_location(dimm, location, sizeof(location));
82 edac_dbg(4, "%s%i: %smapped as virtual row %d, chan %d\n",
83 dimm->mci->mem_is_per_rank ? "rank" : "dimm",
84 number, location, dimm->csrow, dimm->cschannel);
85 edac_dbg(4, " dimm = %p\n", dimm);
86 edac_dbg(4, " dimm->label = '%s'\n", dimm->label);
87 edac_dbg(4, " dimm->nr_pages = 0x%x\n", dimm->nr_pages);
88 edac_dbg(4, " dimm->grain = %d\n", dimm->grain);
89 edac_dbg(4, " dimm->nr_pages = 0x%x\n", dimm->nr_pages);
92 static void edac_mc_dump_csrow(struct csrow_info *csrow)
94 edac_dbg(4, "csrow->csrow_idx = %d\n", csrow->csrow_idx);
95 edac_dbg(4, " csrow = %p\n", csrow);
96 edac_dbg(4, " csrow->first_page = 0x%lx\n", csrow->first_page);
97 edac_dbg(4, " csrow->last_page = 0x%lx\n", csrow->last_page);
98 edac_dbg(4, " csrow->page_mask = 0x%lx\n", csrow->page_mask);
99 edac_dbg(4, " csrow->nr_channels = %d\n", csrow->nr_channels);
100 edac_dbg(4, " csrow->channels = %p\n", csrow->channels);
101 edac_dbg(4, " csrow->mci = %p\n", csrow->mci);
104 static void edac_mc_dump_mci(struct mem_ctl_info *mci)
106 edac_dbg(3, "\tmci = %p\n", mci);
107 edac_dbg(3, "\tmci->mtype_cap = %lx\n", mci->mtype_cap);
108 edac_dbg(3, "\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap);
109 edac_dbg(3, "\tmci->edac_cap = %lx\n", mci->edac_cap);
110 edac_dbg(4, "\tmci->edac_check = %p\n", mci->edac_check);
111 edac_dbg(3, "\tmci->nr_csrows = %d, csrows = %p\n",
112 mci->nr_csrows, mci->csrows);
113 edac_dbg(3, "\tmci->nr_dimms = %d, dimms = %p\n",
114 mci->tot_dimms, mci->dimms);
115 edac_dbg(3, "\tdev = %p\n", mci->pdev);
116 edac_dbg(3, "\tmod_name:ctl_name = %s:%s\n",
117 mci->mod_name, mci->ctl_name);
118 edac_dbg(3, "\tpvt_info = %p\n\n", mci->pvt_info);
121 #endif /* CONFIG_EDAC_DEBUG */
124 * keep those in sync with the enum mem_type
126 const char *edac_mem_types[] = {
128 "Reserved csrow type",
129 "Unknown csrow type",
130 "Fast page mode RAM",
131 "Extended data out RAM",
132 "Burst Extended data out RAM",
133 "Single data rate SDRAM",
134 "Registered single data rate SDRAM",
135 "Double data rate SDRAM",
136 "Registered Double data rate SDRAM",
138 "Unbuffered DDR2 RAM",
139 "Fully buffered DDR2",
140 "Registered DDR2 RAM",
142 "Unbuffered DDR3 RAM",
143 "Registered DDR3 RAM",
145 EXPORT_SYMBOL_GPL(edac_mem_types);
148 * edac_align_ptr - Prepares the pointer offsets for a single-shot allocation
149 * @p: pointer to a pointer with the memory offset to be used. At
150 * return, this will be incremented to point to the next offset
151 * @size: Size of the data structure to be reserved
152 * @n_elems: Number of elements that should be reserved
154 * If 'size' is a constant, the compiler will optimize this whole function
155 * down to either a no-op or the addition of a constant to the value of '*p'.
157 * The 'p' pointer is absolutely needed to keep the proper advancing
158 * further in memory to the proper offsets when allocating the struct along
159 * with its embedded structs, as edac_device_alloc_ctl_info() does it
160 * above, for example.
162 * At return, the pointer 'p' will be incremented to be used on a next call
165 void *edac_align_ptr(void **p, unsigned size, int n_elems)
170 *p += size * n_elems;
173 * 'p' can possibly be an unaligned item X such that sizeof(X) is
174 * 'size'. Adjust 'p' so that its alignment is at least as
175 * stringent as what the compiler would provide for X and return
176 * the aligned result.
177 * Here we assume that the alignment of a "long long" is the most
178 * stringent alignment that the compiler will ever provide by default.
179 * As far as I know, this is a reasonable assumption.
181 if (size > sizeof(long))
182 align = sizeof(long long);
183 else if (size > sizeof(int))
184 align = sizeof(long);
185 else if (size > sizeof(short))
187 else if (size > sizeof(char))
188 align = sizeof(short);
192 r = (unsigned long)p % align;
199 return (void *)(((unsigned long)ptr) + align - r);
202 static void _edac_mc_free(struct mem_ctl_info *mci)
205 struct csrow_info *csr;
206 const unsigned int tot_dimms = mci->tot_dimms;
207 const unsigned int tot_channels = mci->num_cschannel;
208 const unsigned int tot_csrows = mci->nr_csrows;
211 for (i = 0; i < tot_dimms; i++)
212 kfree(mci->dimms[i]);
216 for (row = 0; row < tot_csrows; row++) {
217 csr = mci->csrows[row];
220 for (chn = 0; chn < tot_channels; chn++)
221 kfree(csr->channels[chn]);
222 kfree(csr->channels);
233 * edac_mc_alloc: Allocate and partially fill a struct mem_ctl_info structure
234 * @mc_num: Memory controller number
235 * @n_layers: Number of MC hierarchy layers
236 * layers: Describes each layer as seen by the Memory Controller
237 * @size_pvt: size of private storage needed
240 * Everything is kmalloc'ed as one big chunk - more efficient.
241 * Only can be used if all structures have the same lifetime - otherwise
242 * you have to allocate and initialize your own structures.
244 * Use edac_mc_free() to free mc structures allocated by this function.
246 * NOTE: drivers handle multi-rank memories in different ways: in some
247 * drivers, one multi-rank memory stick is mapped as one entry, while, in
248 * others, a single multi-rank memory stick would be mapped into several
249 * entries. Currently, this function will allocate multiple struct dimm_info
250 * on such scenarios, as grouping the multiple ranks require drivers change.
254 * On success: struct mem_ctl_info pointer
256 struct mem_ctl_info *edac_mc_alloc(unsigned mc_num,
258 struct edac_mc_layer *layers,
261 struct mem_ctl_info *mci;
262 struct edac_mc_layer *layer;
263 struct csrow_info *csr;
264 struct rank_info *chan;
265 struct dimm_info *dimm;
266 u32 *ce_per_layer[EDAC_MAX_LAYERS], *ue_per_layer[EDAC_MAX_LAYERS];
267 unsigned pos[EDAC_MAX_LAYERS];
268 unsigned size, tot_dimms = 1, count = 1;
269 unsigned tot_csrows = 1, tot_channels = 1, tot_errcount = 0;
270 void *pvt, *p, *ptr = NULL;
271 int i, j, row, chn, n, len, off;
272 bool per_rank = false;
274 BUG_ON(n_layers > EDAC_MAX_LAYERS || n_layers == 0);
276 * Calculate the total amount of dimms and csrows/cschannels while
277 * in the old API emulation mode
279 for (i = 0; i < n_layers; i++) {
280 tot_dimms *= layers[i].size;
281 if (layers[i].is_virt_csrow)
282 tot_csrows *= layers[i].size;
284 tot_channels *= layers[i].size;
286 if (layers[i].type == EDAC_MC_LAYER_CHIP_SELECT)
290 /* Figure out the offsets of the various items from the start of an mc
291 * structure. We want the alignment of each item to be at least as
292 * stringent as what the compiler would provide if we could simply
293 * hardcode everything into a single struct.
295 mci = edac_align_ptr(&ptr, sizeof(*mci), 1);
296 layer = edac_align_ptr(&ptr, sizeof(*layer), n_layers);
297 for (i = 0; i < n_layers; i++) {
298 count *= layers[i].size;
299 edac_dbg(4, "errcount layer %d size %d\n", i, count);
300 ce_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count);
301 ue_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count);
302 tot_errcount += 2 * count;
305 edac_dbg(4, "allocating %d error counters\n", tot_errcount);
306 pvt = edac_align_ptr(&ptr, sz_pvt, 1);
307 size = ((unsigned long)pvt) + sz_pvt;
309 edac_dbg(1, "allocating %u bytes for mci data (%d %s, %d csrows/channels)\n",
312 per_rank ? "ranks" : "dimms",
313 tot_csrows * tot_channels);
315 mci = kzalloc(size, GFP_KERNEL);
319 /* Adjust pointers so they point within the memory we just allocated
320 * rather than an imaginary chunk of memory located at address 0.
322 layer = (struct edac_mc_layer *)(((char *)mci) + ((unsigned long)layer));
323 for (i = 0; i < n_layers; i++) {
324 mci->ce_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ce_per_layer[i]));
325 mci->ue_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ue_per_layer[i]));
327 pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL;
329 /* setup index and various internal pointers */
330 mci->mc_idx = mc_num;
331 mci->tot_dimms = tot_dimms;
333 mci->n_layers = n_layers;
335 memcpy(mci->layers, layers, sizeof(*layer) * n_layers);
336 mci->nr_csrows = tot_csrows;
337 mci->num_cschannel = tot_channels;
338 mci->mem_is_per_rank = per_rank;
341 * Alocate and fill the csrow/channels structs
343 mci->csrows = kcalloc(sizeof(*mci->csrows), tot_csrows, GFP_KERNEL);
346 for (row = 0; row < tot_csrows; row++) {
347 csr = kzalloc(sizeof(**mci->csrows), GFP_KERNEL);
350 mci->csrows[row] = csr;
351 csr->csrow_idx = row;
353 csr->nr_channels = tot_channels;
354 csr->channels = kcalloc(sizeof(*csr->channels), tot_channels,
359 for (chn = 0; chn < tot_channels; chn++) {
360 chan = kzalloc(sizeof(**csr->channels), GFP_KERNEL);
363 csr->channels[chn] = chan;
364 chan->chan_idx = chn;
370 * Allocate and fill the dimm structs
372 mci->dimms = kcalloc(sizeof(*mci->dimms), tot_dimms, GFP_KERNEL);
376 memset(&pos, 0, sizeof(pos));
379 for (i = 0; i < tot_dimms; i++) {
380 chan = mci->csrows[row]->channels[chn];
381 off = EDAC_DIMM_OFF(layer, n_layers, pos[0], pos[1], pos[2]);
382 if (off < 0 || off >= tot_dimms) {
383 edac_mc_printk(mci, KERN_ERR, "EDAC core bug: EDAC_DIMM_OFF is trying to do an illegal data access\n");
387 dimm = kzalloc(sizeof(**mci->dimms), GFP_KERNEL);
390 mci->dimms[off] = dimm;
394 * Copy DIMM location and initialize it.
396 len = sizeof(dimm->label);
398 n = snprintf(p, len, "mc#%u", mc_num);
401 for (j = 0; j < n_layers; j++) {
402 n = snprintf(p, len, "%s#%u",
403 edac_layer_name[layers[j].type],
407 dimm->location[j] = pos[j];
413 /* Link it to the csrows old API data */
416 dimm->cschannel = chn;
418 /* Increment csrow location */
420 if (row == tot_csrows) {
425 /* Increment dimm location */
426 for (j = n_layers - 1; j >= 0; j--) {
428 if (pos[j] < layers[j].size)
434 mci->op_state = OP_ALLOC;
436 /* at this point, the root kobj is valid, and in order to
437 * 'free' the object, then the function:
438 * edac_mc_unregister_sysfs_main_kobj() must be called
439 * which will perform kobj unregistration and the actual free
440 * will occur during the kobject callback operation
450 EXPORT_SYMBOL_GPL(edac_mc_alloc);
454 * 'Free' a previously allocated 'mci' structure
455 * @mci: pointer to a struct mem_ctl_info structure
457 void edac_mc_free(struct mem_ctl_info *mci)
461 /* If we're not yet registered with sysfs free only what was allocated
462 * in edac_mc_alloc().
464 if (!device_is_registered(&mci->dev)) {
469 /* the mci instance is freed here, when the sysfs object is dropped */
470 edac_unregister_sysfs(mci);
472 EXPORT_SYMBOL_GPL(edac_mc_free);
478 * scan list of controllers looking for the one that manages
480 * @dev: pointer to a struct device related with the MCI
482 struct mem_ctl_info *find_mci_by_dev(struct device *dev)
484 struct mem_ctl_info *mci;
485 struct list_head *item;
489 list_for_each(item, &mc_devices) {
490 mci = list_entry(item, struct mem_ctl_info, link);
492 if (mci->pdev == dev)
498 EXPORT_SYMBOL_GPL(find_mci_by_dev);
501 * handler for EDAC to check if NMI type handler has asserted interrupt
503 static int edac_mc_assert_error_check_and_clear(void)
507 if (edac_op_state == EDAC_OPSTATE_POLL)
510 old_state = edac_err_assert;
517 * edac_mc_workq_function
518 * performs the operation scheduled by a workq request
520 static void edac_mc_workq_function(struct work_struct *work_req)
522 struct delayed_work *d_work = to_delayed_work(work_req);
523 struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
525 mutex_lock(&mem_ctls_mutex);
527 /* if this control struct has movd to offline state, we are done */
528 if (mci->op_state == OP_OFFLINE) {
529 mutex_unlock(&mem_ctls_mutex);
533 /* Only poll controllers that are running polled and have a check */
534 if (edac_mc_assert_error_check_and_clear() && (mci->edac_check != NULL))
535 mci->edac_check(mci);
537 mutex_unlock(&mem_ctls_mutex);
540 queue_delayed_work(edac_workqueue, &mci->work,
541 msecs_to_jiffies(edac_mc_get_poll_msec()));
545 * edac_mc_workq_setup
546 * initialize a workq item for this mci
547 * passing in the new delay period in msec
551 * called with the mem_ctls_mutex held
553 static void edac_mc_workq_setup(struct mem_ctl_info *mci, unsigned msec)
557 /* if this instance is not in the POLL state, then simply return */
558 if (mci->op_state != OP_RUNNING_POLL)
561 INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
562 queue_delayed_work(edac_workqueue, &mci->work, msecs_to_jiffies(msec));
566 * edac_mc_workq_teardown
567 * stop the workq processing on this mci
571 * called WITHOUT lock held
573 static void edac_mc_workq_teardown(struct mem_ctl_info *mci)
577 if (mci->op_state != OP_RUNNING_POLL)
580 status = cancel_delayed_work(&mci->work);
582 edac_dbg(0, "not canceled, flush the queue\n");
584 /* workq instance might be running, wait for it */
585 flush_workqueue(edac_workqueue);
590 * edac_mc_reset_delay_period(unsigned long value)
592 * user space has updated our poll period value, need to
593 * reset our workq delays
595 void edac_mc_reset_delay_period(int value)
597 struct mem_ctl_info *mci;
598 struct list_head *item;
600 mutex_lock(&mem_ctls_mutex);
602 /* scan the list and turn off all workq timers, doing so under lock
604 list_for_each(item, &mc_devices) {
605 mci = list_entry(item, struct mem_ctl_info, link);
607 if (mci->op_state == OP_RUNNING_POLL)
608 cancel_delayed_work(&mci->work);
611 mutex_unlock(&mem_ctls_mutex);
614 /* re-walk the list, and reset the poll delay */
615 mutex_lock(&mem_ctls_mutex);
617 list_for_each(item, &mc_devices) {
618 mci = list_entry(item, struct mem_ctl_info, link);
620 edac_mc_workq_setup(mci, (unsigned long) value);
623 mutex_unlock(&mem_ctls_mutex);
628 /* Return 0 on success, 1 on failure.
629 * Before calling this function, caller must
630 * assign a unique value to mci->mc_idx.
634 * called with the mem_ctls_mutex lock held
636 static int add_mc_to_global_list(struct mem_ctl_info *mci)
638 struct list_head *item, *insert_before;
639 struct mem_ctl_info *p;
641 insert_before = &mc_devices;
643 p = find_mci_by_dev(mci->pdev);
644 if (unlikely(p != NULL))
647 list_for_each(item, &mc_devices) {
648 p = list_entry(item, struct mem_ctl_info, link);
650 if (p->mc_idx >= mci->mc_idx) {
651 if (unlikely(p->mc_idx == mci->mc_idx))
654 insert_before = item;
659 list_add_tail_rcu(&mci->link, insert_before);
660 atomic_inc(&edac_handlers);
664 edac_printk(KERN_WARNING, EDAC_MC,
665 "%s (%s) %s %s already assigned %d\n", dev_name(p->pdev),
666 edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
670 edac_printk(KERN_WARNING, EDAC_MC,
671 "bug in low-level driver: attempt to assign\n"
672 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
676 static void del_mc_from_global_list(struct mem_ctl_info *mci)
678 atomic_dec(&edac_handlers);
679 list_del_rcu(&mci->link);
681 /* these are for safe removal of devices from global list while
682 * NMI handlers may be traversing list
685 INIT_LIST_HEAD(&mci->link);
689 * edac_mc_find: Search for a mem_ctl_info structure whose index is 'idx'.
691 * If found, return a pointer to the structure.
694 * Caller must hold mem_ctls_mutex.
696 struct mem_ctl_info *edac_mc_find(int idx)
698 struct list_head *item;
699 struct mem_ctl_info *mci;
701 list_for_each(item, &mc_devices) {
702 mci = list_entry(item, struct mem_ctl_info, link);
704 if (mci->mc_idx >= idx) {
705 if (mci->mc_idx == idx)
714 EXPORT_SYMBOL(edac_mc_find);
717 * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and
718 * create sysfs entries associated with mci structure
719 * @mci: pointer to the mci structure to be added to the list
726 /* FIXME - should a warning be printed if no error detection? correction? */
727 int edac_mc_add_mc(struct mem_ctl_info *mci)
731 #ifdef CONFIG_EDAC_DEBUG
732 if (edac_debug_level >= 3)
733 edac_mc_dump_mci(mci);
735 if (edac_debug_level >= 4) {
738 for (i = 0; i < mci->nr_csrows; i++) {
739 struct csrow_info *csrow = mci->csrows[i];
743 for (j = 0; j < csrow->nr_channels; j++)
744 nr_pages += csrow->channels[j]->dimm->nr_pages;
747 edac_mc_dump_csrow(csrow);
748 for (j = 0; j < csrow->nr_channels; j++)
749 if (csrow->channels[j]->dimm->nr_pages)
750 edac_mc_dump_channel(csrow->channels[j]);
752 for (i = 0; i < mci->tot_dimms; i++)
753 if (mci->dimms[i]->nr_pages)
754 edac_mc_dump_dimm(mci->dimms[i], i);
757 mutex_lock(&mem_ctls_mutex);
759 if (add_mc_to_global_list(mci))
762 /* set load time so that error rate can be tracked */
763 mci->start_time = jiffies;
765 if (edac_create_sysfs_mci_device(mci)) {
766 edac_mc_printk(mci, KERN_WARNING,
767 "failed to create sysfs device\n");
771 /* If there IS a check routine, then we are running POLLED */
772 if (mci->edac_check != NULL) {
773 /* This instance is NOW RUNNING */
774 mci->op_state = OP_RUNNING_POLL;
776 edac_mc_workq_setup(mci, edac_mc_get_poll_msec());
778 mci->op_state = OP_RUNNING_INTERRUPT;
781 /* Report action taken */
782 edac_mc_printk(mci, KERN_INFO, "Giving out device to '%s' '%s':"
783 " DEV %s\n", mci->mod_name, mci->ctl_name, edac_dev_name(mci));
785 mutex_unlock(&mem_ctls_mutex);
789 del_mc_from_global_list(mci);
792 mutex_unlock(&mem_ctls_mutex);
795 EXPORT_SYMBOL_GPL(edac_mc_add_mc);
798 * edac_mc_del_mc: Remove sysfs entries for specified mci structure and
799 * remove mci structure from global list
800 * @pdev: Pointer to 'struct device' representing mci structure to remove.
802 * Return pointer to removed mci structure, or NULL if device not found.
804 struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
806 struct mem_ctl_info *mci;
810 mutex_lock(&mem_ctls_mutex);
812 /* find the requested mci struct in the global list */
813 mci = find_mci_by_dev(dev);
815 mutex_unlock(&mem_ctls_mutex);
819 del_mc_from_global_list(mci);
820 mutex_unlock(&mem_ctls_mutex);
822 /* flush workq processes */
823 edac_mc_workq_teardown(mci);
825 /* marking MCI offline */
826 mci->op_state = OP_OFFLINE;
828 /* remove from sysfs */
829 edac_remove_sysfs_mci_device(mci);
831 edac_printk(KERN_INFO, EDAC_MC,
832 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
833 mci->mod_name, mci->ctl_name, edac_dev_name(mci));
837 EXPORT_SYMBOL_GPL(edac_mc_del_mc);
839 static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
844 unsigned long flags = 0;
848 /* ECC error page was not in our memory. Ignore it. */
849 if (!pfn_valid(page))
852 /* Find the actual page structure then map it and fix */
853 pg = pfn_to_page(page);
856 local_irq_save(flags);
858 virt_addr = kmap_atomic(pg);
860 /* Perform architecture specific atomic scrub operation */
861 atomic_scrub(virt_addr + offset, size);
863 /* Unmap and complete */
864 kunmap_atomic(virt_addr);
867 local_irq_restore(flags);
870 /* FIXME - should return -1 */
871 int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
873 struct csrow_info **csrows = mci->csrows;
876 edac_dbg(1, "MC%d: 0x%lx\n", mci->mc_idx, page);
879 for (i = 0; i < mci->nr_csrows; i++) {
880 struct csrow_info *csrow = csrows[i];
882 for (j = 0; j < csrow->nr_channels; j++) {
883 struct dimm_info *dimm = csrow->channels[j]->dimm;
889 edac_dbg(3, "MC%d: first(0x%lx) page(0x%lx) last(0x%lx) mask(0x%lx)\n",
891 csrow->first_page, page, csrow->last_page,
894 if ((page >= csrow->first_page) &&
895 (page <= csrow->last_page) &&
896 ((page & csrow->page_mask) ==
897 (csrow->first_page & csrow->page_mask))) {
904 edac_mc_printk(mci, KERN_ERR,
905 "could not look up page error address %lx\n",
906 (unsigned long)page);
910 EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
912 const char *edac_layer_name[] = {
913 [EDAC_MC_LAYER_BRANCH] = "branch",
914 [EDAC_MC_LAYER_CHANNEL] = "channel",
915 [EDAC_MC_LAYER_SLOT] = "slot",
916 [EDAC_MC_LAYER_CHIP_SELECT] = "csrow",
918 EXPORT_SYMBOL_GPL(edac_layer_name);
920 static void edac_inc_ce_error(struct mem_ctl_info *mci,
921 bool enable_per_layer_report,
922 const int pos[EDAC_MAX_LAYERS],
929 if (!enable_per_layer_report) {
930 mci->ce_noinfo_count += count;
934 for (i = 0; i < mci->n_layers; i++) {
938 mci->ce_per_layer[i][index] += count;
940 if (i < mci->n_layers - 1)
941 index *= mci->layers[i + 1].size;
945 static void edac_inc_ue_error(struct mem_ctl_info *mci,
946 bool enable_per_layer_report,
947 const int pos[EDAC_MAX_LAYERS],
954 if (!enable_per_layer_report) {
955 mci->ce_noinfo_count += count;
959 for (i = 0; i < mci->n_layers; i++) {
963 mci->ue_per_layer[i][index] += count;
965 if (i < mci->n_layers - 1)
966 index *= mci->layers[i + 1].size;
970 static void edac_ce_error(struct mem_ctl_info *mci,
971 const u16 error_count,
972 const int pos[EDAC_MAX_LAYERS],
974 const char *location,
977 const char *other_detail,
978 const bool enable_per_layer_report,
979 const unsigned long page_frame_number,
980 const unsigned long offset_in_page,
983 unsigned long remapped_page;
985 if (edac_mc_get_log_ce()) {
986 if (other_detail && *other_detail)
987 edac_mc_printk(mci, KERN_WARNING,
988 "%d CE %s on %s (%s %s - %s)\n",
990 msg, label, location,
991 detail, other_detail);
993 edac_mc_printk(mci, KERN_WARNING,
994 "%d CE %s on %s (%s %s)\n",
996 msg, label, location,
999 edac_inc_ce_error(mci, enable_per_layer_report, pos, error_count);
1001 if (mci->scrub_mode & SCRUB_SW_SRC) {
1003 * Some memory controllers (called MCs below) can remap
1004 * memory so that it is still available at a different
1005 * address when PCI devices map into memory.
1006 * MC's that can't do this, lose the memory where PCI
1007 * devices are mapped. This mapping is MC-dependent
1008 * and so we call back into the MC driver for it to
1009 * map the MC page to a physical (CPU) page which can
1010 * then be mapped to a virtual page - which can then
1013 remapped_page = mci->ctl_page_to_phys ?
1014 mci->ctl_page_to_phys(mci, page_frame_number) :
1017 edac_mc_scrub_block(remapped_page,
1018 offset_in_page, grain);
1022 static void edac_ue_error(struct mem_ctl_info *mci,
1023 const u16 error_count,
1024 const int pos[EDAC_MAX_LAYERS],
1026 const char *location,
1029 const char *other_detail,
1030 const bool enable_per_layer_report)
1032 if (edac_mc_get_log_ue()) {
1033 if (other_detail && *other_detail)
1034 edac_mc_printk(mci, KERN_WARNING,
1035 "%d UE %s on %s (%s %s - %s)\n",
1037 msg, label, location, detail,
1040 edac_mc_printk(mci, KERN_WARNING,
1041 "%d UE %s on %s (%s %s)\n",
1043 msg, label, location, detail);
1046 if (edac_mc_get_panic_on_ue()) {
1047 if (other_detail && *other_detail)
1048 panic("UE %s on %s (%s%s - %s)\n",
1049 msg, label, location, detail, other_detail);
1051 panic("UE %s on %s (%s%s)\n",
1052 msg, label, location, detail);
1055 edac_inc_ue_error(mci, enable_per_layer_report, pos, error_count);
1058 #define OTHER_LABEL " or "
1061 * edac_mc_handle_error - reports a memory event to userspace
1063 * @type: severity of the error (CE/UE/Fatal)
1064 * @mci: a struct mem_ctl_info pointer
1065 * @error_count: Number of errors of the same type
1066 * @page_frame_number: mem page where the error occurred
1067 * @offset_in_page: offset of the error inside the page
1068 * @syndrome: ECC syndrome
1069 * @top_layer: Memory layer[0] position
1070 * @mid_layer: Memory layer[1] position
1071 * @low_layer: Memory layer[2] position
1072 * @msg: Message meaningful to the end users that
1073 * explains the event
1074 * @other_detail: Technical details about the event that
1075 * may help hardware manufacturers and
1076 * EDAC developers to analyse the event
1078 void edac_mc_handle_error(const enum hw_event_mc_err_type type,
1079 struct mem_ctl_info *mci,
1080 const u16 error_count,
1081 const unsigned long page_frame_number,
1082 const unsigned long offset_in_page,
1083 const unsigned long syndrome,
1084 const int top_layer,
1085 const int mid_layer,
1086 const int low_layer,
1088 const char *other_detail)
1090 /* FIXME: too much for stack: move it to some pre-alocated area */
1091 char detail[80], location[80];
1092 char label[(EDAC_MC_LABEL_LEN + 1 + sizeof(OTHER_LABEL)) * mci->tot_dimms];
1094 int row = -1, chan = -1;
1095 int pos[EDAC_MAX_LAYERS] = { top_layer, mid_layer, low_layer };
1098 bool enable_per_layer_report = false;
1101 edac_dbg(3, "MC%d\n", mci->mc_idx);
1104 * Check if the event report is consistent and if the memory
1105 * location is known. If it is known, enable_per_layer_report will be
1106 * true, the DIMM(s) label info will be filled and the per-layer
1107 * error counters will be incremented.
1109 for (i = 0; i < mci->n_layers; i++) {
1110 if (pos[i] >= (int)mci->layers[i].size) {
1111 if (type == HW_EVENT_ERR_CORRECTED)
1116 edac_mc_printk(mci, KERN_ERR,
1117 "INTERNAL ERROR: %s value is out of range (%d >= %d)\n",
1118 edac_layer_name[mci->layers[i].type],
1119 pos[i], mci->layers[i].size);
1121 * Instead of just returning it, let's use what's
1122 * known about the error. The increment routines and
1123 * the DIMM filter logic will do the right thing by
1124 * pointing the likely damaged DIMMs.
1129 enable_per_layer_report = true;
1133 * Get the dimm label/grain that applies to the match criteria.
1134 * As the error algorithm may not be able to point to just one memory
1135 * stick, the logic here will get all possible labels that could
1136 * pottentially be affected by the error.
1137 * On FB-DIMM memory controllers, for uncorrected errors, it is common
1138 * to have only the MC channel and the MC dimm (also called "branch")
1139 * but the channel is not known, as the memory is arranged in pairs,
1140 * where each memory belongs to a separate channel within the same
1146 for (i = 0; i < mci->tot_dimms; i++) {
1147 struct dimm_info *dimm = mci->dimms[i];
1149 if (top_layer >= 0 && top_layer != dimm->location[0])
1151 if (mid_layer >= 0 && mid_layer != dimm->location[1])
1153 if (low_layer >= 0 && low_layer != dimm->location[2])
1156 /* get the max grain, over the error match range */
1157 if (dimm->grain > grain)
1158 grain = dimm->grain;
1161 * If the error is memory-controller wide, there's no need to
1162 * seek for the affected DIMMs because the whole
1163 * channel/memory controller/... may be affected.
1164 * Also, don't show errors for empty DIMM slots.
1166 if (enable_per_layer_report && dimm->nr_pages) {
1168 strcpy(p, OTHER_LABEL);
1169 p += strlen(OTHER_LABEL);
1171 strcpy(p, dimm->label);
1176 * get csrow/channel of the DIMM, in order to allow
1177 * incrementing the compat API counters
1179 edac_dbg(4, "%s csrows map: (%d,%d)\n",
1180 mci->mem_is_per_rank ? "rank" : "dimm",
1181 dimm->csrow, dimm->cschannel);
1184 else if (row >= 0 && row != dimm->csrow)
1188 chan = dimm->cschannel;
1189 else if (chan >= 0 && chan != dimm->cschannel)
1194 if (!enable_per_layer_report) {
1195 strcpy(label, "any memory");
1197 edac_dbg(4, "csrow/channel to increment: (%d,%d)\n", row, chan);
1199 strcpy(label, "unknown memory");
1200 if (type == HW_EVENT_ERR_CORRECTED) {
1202 mci->csrows[row]->ce_count += error_count;
1204 mci->csrows[row]->channels[chan]->ce_count += error_count;
1208 mci->csrows[row]->ue_count += error_count;
1211 /* Fill the RAM location data */
1213 for (i = 0; i < mci->n_layers; i++) {
1217 p += sprintf(p, "%s:%d ",
1218 edac_layer_name[mci->layers[i].type],
1224 /* Report the error via the trace interface */
1226 grain_bits = fls_long(grain) + 1;
1227 trace_mc_event(type, msg, label, error_count,
1228 mci->mc_idx, top_layer, mid_layer, low_layer,
1229 PAGES_TO_MiB(page_frame_number) | offset_in_page,
1230 grain_bits, syndrome, other_detail);
1232 /* Memory type dependent details about the error */
1233 if (type == HW_EVENT_ERR_CORRECTED) {
1234 snprintf(detail, sizeof(detail),
1235 "page:0x%lx offset:0x%lx grain:%ld syndrome:0x%lx",
1236 page_frame_number, offset_in_page,
1238 edac_ce_error(mci, error_count, pos, msg, location, label,
1239 detail, other_detail, enable_per_layer_report,
1240 page_frame_number, offset_in_page, grain);
1242 snprintf(detail, sizeof(detail),
1243 "page:0x%lx offset:0x%lx grain:%ld",
1244 page_frame_number, offset_in_page, grain);
1246 edac_ue_error(mci, error_count, pos, msg, location, label,
1247 detail, other_detail, enable_per_layer_report);
1250 EXPORT_SYMBOL_GPL(edac_mc_handle_error);