2 * edac_mc kernel module
3 * (C) 2005 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/sysdev.h>
29 #include <linux/ctype.h>
30 #include <linux/kthread.h>
31 #include <asm/uaccess.h>
36 #define EDAC_MC_VERSION "Ver: 2.0.0 " __DATE__
38 /* For now, disable the EDAC sysfs code. The sysfs interface that EDAC
39 * presents to user space needs more thought, and is likely to change
42 #define DISABLE_EDAC_SYSFS
44 #ifdef CONFIG_EDAC_DEBUG
45 /* Values of 0 to 4 will generate output */
46 int edac_debug_level = 1;
47 EXPORT_SYMBOL_GPL(edac_debug_level);
50 /* EDAC Controls, setable by module parameter, and sysfs */
51 static int log_ue = 1;
52 static int log_ce = 1;
53 static int panic_on_ue;
54 static int poll_msec = 1000;
56 /* lock to memory controller's control array */
57 static DECLARE_MUTEX(mem_ctls_mutex);
58 static struct list_head mc_devices = LIST_HEAD_INIT(mc_devices);
60 static struct task_struct *edac_thread;
63 static int check_pci_parity = 0; /* default YES check PCI parity */
64 static int panic_on_pci_parity; /* default no panic on PCI Parity */
65 static atomic_t pci_parity_count = ATOMIC_INIT(0);
67 /* Structure of the whitelist and blacklist arrays */
68 struct edac_pci_device_list {
69 unsigned int vendor; /* Vendor ID */
70 unsigned int device; /* Deviice ID */
73 #define MAX_LISTED_PCI_DEVICES 32
75 /* List of PCI devices (vendor-id:device-id) that should be skipped */
76 static struct edac_pci_device_list pci_blacklist[MAX_LISTED_PCI_DEVICES];
77 static int pci_blacklist_count;
79 /* List of PCI devices (vendor-id:device-id) that should be scanned */
80 static struct edac_pci_device_list pci_whitelist[MAX_LISTED_PCI_DEVICES];
81 static int pci_whitelist_count ;
83 #ifndef DISABLE_EDAC_SYSFS
84 static struct kobject edac_pci_kobj; /* /sys/devices/system/edac/pci */
85 static struct completion edac_pci_kobj_complete;
86 #endif /* DISABLE_EDAC_SYSFS */
87 #endif /* CONFIG_PCI */
89 /* START sysfs data and methods */
91 #ifndef DISABLE_EDAC_SYSFS
93 static const char *mem_types[] = {
94 [MEM_EMPTY] = "Empty",
95 [MEM_RESERVED] = "Reserved",
96 [MEM_UNKNOWN] = "Unknown",
100 [MEM_SDR] = "Unbuffered-SDR",
101 [MEM_RDR] = "Registered-SDR",
102 [MEM_DDR] = "Unbuffered-DDR",
103 [MEM_RDDR] = "Registered-DDR",
107 static const char *dev_types[] = {
108 [DEV_UNKNOWN] = "Unknown",
118 static const char *edac_caps[] = {
119 [EDAC_UNKNOWN] = "Unknown",
120 [EDAC_NONE] = "None",
121 [EDAC_RESERVED] = "Reserved",
122 [EDAC_PARITY] = "PARITY",
124 [EDAC_SECDED] = "SECDED",
125 [EDAC_S2ECD2ED] = "S2ECD2ED",
126 [EDAC_S4ECD4ED] = "S4ECD4ED",
127 [EDAC_S8ECD8ED] = "S8ECD8ED",
128 [EDAC_S16ECD16ED] = "S16ECD16ED"
131 /* sysfs object: /sys/devices/system/edac */
132 static struct sysdev_class edac_class = {
133 set_kset_name("edac"),
137 * /sys/devices/system/edac/mc
139 static struct kobject edac_memctrl_kobj;
141 /* We use these to wait for the reference counts on edac_memctrl_kobj and
142 * edac_pci_kobj to reach 0.
144 static struct completion edac_memctrl_kobj_complete;
147 * /sys/devices/system/edac/mc;
148 * data structures and methods
151 static ssize_t memctrl_string_show(void *ptr, char *buffer)
153 char *value = (char*) ptr;
154 return sprintf(buffer, "%s\n", value);
158 static ssize_t memctrl_int_show(void *ptr, char *buffer)
160 int *value = (int*) ptr;
161 return sprintf(buffer, "%d\n", *value);
164 static ssize_t memctrl_int_store(void *ptr, const char *buffer, size_t count)
166 int *value = (int*) ptr;
168 if (isdigit(*buffer))
169 *value = simple_strtoul(buffer, NULL, 0);
174 struct memctrl_dev_attribute {
175 struct attribute attr;
177 ssize_t (*show)(void *,char *);
178 ssize_t (*store)(void *, const char *, size_t);
181 /* Set of show/store abstract level functions for memory control object */
182 static ssize_t memctrl_dev_show(struct kobject *kobj,
183 struct attribute *attr, char *buffer)
185 struct memctrl_dev_attribute *memctrl_dev;
186 memctrl_dev = (struct memctrl_dev_attribute*)attr;
188 if (memctrl_dev->show)
189 return memctrl_dev->show(memctrl_dev->value, buffer);
194 static ssize_t memctrl_dev_store(struct kobject *kobj, struct attribute *attr,
195 const char *buffer, size_t count)
197 struct memctrl_dev_attribute *memctrl_dev;
198 memctrl_dev = (struct memctrl_dev_attribute*)attr;
200 if (memctrl_dev->store)
201 return memctrl_dev->store(memctrl_dev->value, buffer, count);
206 static struct sysfs_ops memctrlfs_ops = {
207 .show = memctrl_dev_show,
208 .store = memctrl_dev_store
211 #define MEMCTRL_ATTR(_name,_mode,_show,_store) \
212 struct memctrl_dev_attribute attr_##_name = { \
213 .attr = {.name = __stringify(_name), .mode = _mode }, \
219 #define MEMCTRL_STRING_ATTR(_name,_data,_mode,_show,_store) \
220 struct memctrl_dev_attribute attr_##_name = { \
221 .attr = {.name = __stringify(_name), .mode = _mode }, \
227 /* cwrow<id> attribute f*/
229 MEMCTRL_STRING_ATTR(mc_version,EDAC_MC_VERSION,S_IRUGO,memctrl_string_show,NULL);
232 /* csrow<id> control files */
233 MEMCTRL_ATTR(panic_on_ue,S_IRUGO|S_IWUSR,memctrl_int_show,memctrl_int_store);
234 MEMCTRL_ATTR(log_ue,S_IRUGO|S_IWUSR,memctrl_int_show,memctrl_int_store);
235 MEMCTRL_ATTR(log_ce,S_IRUGO|S_IWUSR,memctrl_int_show,memctrl_int_store);
236 MEMCTRL_ATTR(poll_msec,S_IRUGO|S_IWUSR,memctrl_int_show,memctrl_int_store);
238 /* Base Attributes of the memory ECC object */
239 static struct memctrl_dev_attribute *memctrl_attr[] = {
247 /* Main MC kobject release() function */
248 static void edac_memctrl_master_release(struct kobject *kobj)
250 debugf1("%s()\n", __func__);
251 complete(&edac_memctrl_kobj_complete);
254 static struct kobj_type ktype_memctrl = {
255 .release = edac_memctrl_master_release,
256 .sysfs_ops = &memctrlfs_ops,
257 .default_attrs = (struct attribute **) memctrl_attr,
260 #endif /* DISABLE_EDAC_SYSFS */
262 /* Initialize the main sysfs entries for edac:
263 * /sys/devices/system/edac
270 static int edac_sysfs_memctrl_setup(void)
271 #ifdef DISABLE_EDAC_SYSFS
279 debugf1("%s()\n", __func__);
281 /* create the /sys/devices/system/edac directory */
282 err = sysdev_class_register(&edac_class);
285 /* Init the MC's kobject */
286 memset(&edac_memctrl_kobj, 0, sizeof (edac_memctrl_kobj));
287 edac_memctrl_kobj.parent = &edac_class.kset.kobj;
288 edac_memctrl_kobj.ktype = &ktype_memctrl;
290 /* generate sysfs "..../edac/mc" */
291 err = kobject_set_name(&edac_memctrl_kobj,"mc");
294 /* FIXME: maybe new sysdev_create_subdir() */
295 err = kobject_register(&edac_memctrl_kobj);
298 debugf1("Failed to register '.../edac/mc'\n");
300 debugf1("Registered '.../edac/mc' kobject\n");
303 debugf1("%s() error=%d\n", __func__, err);
307 #endif /* DISABLE_EDAC_SYSFS */
311 * the '..../edac/mc' kobject followed by '..../edac' itself
313 static void edac_sysfs_memctrl_teardown(void)
315 #ifndef DISABLE_EDAC_SYSFS
316 debugf0("MC: " __FILE__ ": %s()\n", __func__);
318 /* Unregister the MC's kobject and wait for reference count to reach
321 init_completion(&edac_memctrl_kobj_complete);
322 kobject_unregister(&edac_memctrl_kobj);
323 wait_for_completion(&edac_memctrl_kobj_complete);
325 /* Unregister the 'edac' object */
326 sysdev_class_unregister(&edac_class);
327 #endif /* DISABLE_EDAC_SYSFS */
332 #ifndef DISABLE_EDAC_SYSFS
335 * /sys/devices/system/edac/pci;
336 * data structures and methods
339 struct list_control {
340 struct edac_pci_device_list *list;
345 /* Output the list as: vendor_id:device:id<,vendor_id:device_id> */
346 static ssize_t edac_pci_list_string_show(void *ptr, char *buffer)
348 struct list_control *listctl;
349 struct edac_pci_device_list *list;
355 list = listctl->list;
357 for (i = 0; i < *(listctl->count); i++, list++ ) {
359 len += snprintf(p + len, (PAGE_SIZE-len), ",");
361 len += snprintf(p + len,
364 list->vendor,list->device);
367 len += snprintf(p + len,(PAGE_SIZE-len), "\n");
368 return (ssize_t) len;
373 * Scan string from **s to **e looking for one 'vendor:device' tuple
374 * where each field is a hex value
376 * return 0 if an entry is NOT found
377 * return 1 if an entry is found
378 * fill in *vendor_id and *device_id with values found
380 * In both cases, make sure *s has been moved forward toward *e
382 static int parse_one_device(const char **s,const char **e,
383 unsigned int *vendor_id, unsigned int *device_id)
385 const char *runner, *p;
387 /* if null byte, we are done */
389 (*s)++; /* keep *s moving */
393 /* skip over newlines & whitespace */
394 if ((**s == '\n') || isspace(**s)) {
399 if (!isxdigit(**s)) {
404 /* parse vendor_id */
407 while (runner < *e) {
408 /* scan for vendor:device delimiter */
409 if (*runner == ':') {
410 *vendor_id = simple_strtol((char*) *s, (char**) &p, 16);
418 if (!isxdigit(*runner)) {
423 /* parse device_id */
425 *device_id = simple_strtol((char*)runner, (char**)&p, 16);
433 static ssize_t edac_pci_list_string_store(void *ptr, const char *buffer,
436 struct list_control *listctl;
437 struct edac_pci_device_list *list;
438 unsigned int vendor_id, device_id;
445 list = listctl->list;
446 index = listctl->count;
449 while (*index < MAX_LISTED_PCI_DEVICES) {
450 if (parse_one_device(&s,&e,&vendor_id,&device_id)) {
451 list[ *index ].vendor = vendor_id;
452 list[ *index ].device = device_id;
456 /* check for all data consume */
465 static ssize_t edac_pci_int_show(void *ptr, char *buffer)
468 return sprintf(buffer,"%d\n",*value);
471 static ssize_t edac_pci_int_store(void *ptr, const char *buffer, size_t count)
475 if (isdigit(*buffer))
476 *value = simple_strtoul(buffer,NULL,0);
481 struct edac_pci_dev_attribute {
482 struct attribute attr;
484 ssize_t (*show)(void *,char *);
485 ssize_t (*store)(void *, const char *,size_t);
488 /* Set of show/store abstract level functions for PCI Parity object */
489 static ssize_t edac_pci_dev_show(struct kobject *kobj, struct attribute *attr,
492 struct edac_pci_dev_attribute *edac_pci_dev;
493 edac_pci_dev= (struct edac_pci_dev_attribute*)attr;
495 if (edac_pci_dev->show)
496 return edac_pci_dev->show(edac_pci_dev->value, buffer);
500 static ssize_t edac_pci_dev_store(struct kobject *kobj,
501 struct attribute *attr, const char *buffer, size_t count)
503 struct edac_pci_dev_attribute *edac_pci_dev;
504 edac_pci_dev= (struct edac_pci_dev_attribute*)attr;
506 if (edac_pci_dev->show)
507 return edac_pci_dev->store(edac_pci_dev->value, buffer, count);
511 static struct sysfs_ops edac_pci_sysfs_ops = {
512 .show = edac_pci_dev_show,
513 .store = edac_pci_dev_store
516 #define EDAC_PCI_ATTR(_name,_mode,_show,_store) \
517 struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
518 .attr = {.name = __stringify(_name), .mode = _mode }, \
524 #define EDAC_PCI_STRING_ATTR(_name,_data,_mode,_show,_store) \
525 struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
526 .attr = {.name = __stringify(_name), .mode = _mode }, \
533 static struct list_control pci_whitelist_control = {
534 .list = pci_whitelist,
535 .count = &pci_whitelist_count
538 static struct list_control pci_blacklist_control = {
539 .list = pci_blacklist,
540 .count = &pci_blacklist_count
543 /* whitelist attribute */
544 EDAC_PCI_STRING_ATTR(pci_parity_whitelist,
545 &pci_whitelist_control,
547 edac_pci_list_string_show,
548 edac_pci_list_string_store);
550 EDAC_PCI_STRING_ATTR(pci_parity_blacklist,
551 &pci_blacklist_control,
553 edac_pci_list_string_show,
554 edac_pci_list_string_store);
557 /* PCI Parity control files */
558 EDAC_PCI_ATTR(check_pci_parity, S_IRUGO|S_IWUSR, edac_pci_int_show,
560 EDAC_PCI_ATTR(panic_on_pci_parity, S_IRUGO|S_IWUSR, edac_pci_int_show,
562 EDAC_PCI_ATTR(pci_parity_count, S_IRUGO, edac_pci_int_show, NULL);
564 /* Base Attributes of the memory ECC object */
565 static struct edac_pci_dev_attribute *edac_pci_attr[] = {
566 &edac_pci_attr_check_pci_parity,
567 &edac_pci_attr_panic_on_pci_parity,
568 &edac_pci_attr_pci_parity_count,
572 /* No memory to release */
573 static void edac_pci_release(struct kobject *kobj)
575 debugf1("%s()\n", __func__);
576 complete(&edac_pci_kobj_complete);
579 static struct kobj_type ktype_edac_pci = {
580 .release = edac_pci_release,
581 .sysfs_ops = &edac_pci_sysfs_ops,
582 .default_attrs = (struct attribute **) edac_pci_attr,
585 #endif /* DISABLE_EDAC_SYSFS */
588 * edac_sysfs_pci_setup()
591 static int edac_sysfs_pci_setup(void)
592 #ifdef DISABLE_EDAC_SYSFS
600 debugf1("%s()\n", __func__);
602 memset(&edac_pci_kobj, 0, sizeof(edac_pci_kobj));
603 edac_pci_kobj.parent = &edac_class.kset.kobj;
604 edac_pci_kobj.ktype = &ktype_edac_pci;
605 err = kobject_set_name(&edac_pci_kobj, "pci");
608 /* Instanstiate the csrow object */
609 /* FIXME: maybe new sysdev_create_subdir() */
610 err = kobject_register(&edac_pci_kobj);
613 debugf1("Failed to register '.../edac/pci'\n");
615 debugf1("Registered '.../edac/pci' kobject\n");
620 #endif /* DISABLE_EDAC_SYSFS */
622 static void edac_sysfs_pci_teardown(void)
624 #ifndef DISABLE_EDAC_SYSFS
625 debugf0("%s()\n", __func__);
626 init_completion(&edac_pci_kobj_complete);
627 kobject_unregister(&edac_pci_kobj);
628 wait_for_completion(&edac_pci_kobj_complete);
633 static u16 get_pci_parity_status(struct pci_dev *dev, int secondary)
638 where = secondary ? PCI_SEC_STATUS : PCI_STATUS;
639 pci_read_config_word(dev, where, &status);
641 /* If we get back 0xFFFF then we must suspect that the card has been
642 * pulled but the Linux PCI layer has not yet finished cleaning up.
643 * We don't want to report on such devices
646 if (status == 0xFFFF) {
649 pci_read_config_dword(dev, 0, &sanity);
651 if (sanity == 0xFFFFFFFF)
655 status &= PCI_STATUS_DETECTED_PARITY | PCI_STATUS_SIG_SYSTEM_ERROR |
659 /* reset only the bits we are interested in */
660 pci_write_config_word(dev, where, status);
665 typedef void (*pci_parity_check_fn_t) (struct pci_dev *dev);
667 /* Clear any PCI parity errors logged by this device. */
668 static void edac_pci_dev_parity_clear(struct pci_dev *dev)
672 get_pci_parity_status(dev, 0);
674 /* read the device TYPE, looking for bridges */
675 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
677 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE)
678 get_pci_parity_status(dev, 1);
685 static void edac_pci_dev_parity_test(struct pci_dev *dev)
690 /* read the STATUS register on this device
692 status = get_pci_parity_status(dev, 0);
694 debugf2("PCI STATUS= 0x%04x %s\n", status, dev->dev.bus_id );
696 /* check the status reg for errors */
698 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR))
699 edac_printk(KERN_CRIT, EDAC_PCI,
700 "Signaled System Error on %s\n",
703 if (status & (PCI_STATUS_PARITY)) {
704 edac_printk(KERN_CRIT, EDAC_PCI,
705 "Master Data Parity Error on %s\n",
708 atomic_inc(&pci_parity_count);
711 if (status & (PCI_STATUS_DETECTED_PARITY)) {
712 edac_printk(KERN_CRIT, EDAC_PCI,
713 "Detected Parity Error on %s\n",
716 atomic_inc(&pci_parity_count);
720 /* read the device TYPE, looking for bridges */
721 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
723 debugf2("PCI HEADER TYPE= 0x%02x %s\n", header_type, dev->dev.bus_id );
725 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) {
726 /* On bridges, need to examine secondary status register */
727 status = get_pci_parity_status(dev, 1);
729 debugf2("PCI SEC_STATUS= 0x%04x %s\n",
730 status, dev->dev.bus_id );
732 /* check the secondary status reg for errors */
734 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR))
735 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
736 "Signaled System Error on %s\n",
739 if (status & (PCI_STATUS_PARITY)) {
740 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
741 "Master Data Parity Error on "
742 "%s\n", pci_name(dev));
744 atomic_inc(&pci_parity_count);
747 if (status & (PCI_STATUS_DETECTED_PARITY)) {
748 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
749 "Detected Parity Error on %s\n",
752 atomic_inc(&pci_parity_count);
759 * check_dev_on_list: Scan for a PCI device on a white/black list
760 * @list: an EDAC &edac_pci_device_list white/black list pointer
761 * @free_index: index of next free entry on the list
762 * @pci_dev: PCI Device pointer
764 * see if list contains the device.
766 * Returns: 0 not found
769 static int check_dev_on_list(struct edac_pci_device_list *list,
770 int free_index, struct pci_dev *dev)
773 int rc = 0; /* Assume not found */
774 unsigned short vendor=dev->vendor;
775 unsigned short device=dev->device;
777 /* Scan the list, looking for a vendor/device match */
778 for (i = 0; i < free_index; i++, list++ ) {
779 if ((list->vendor == vendor ) && (list->device == device )) {
789 * pci_dev parity list iterator
790 * Scan the PCI device list for one iteration, looking for SERRORs
791 * Master Parity ERRORS or Parity ERRORs on primary or secondary devices
793 static inline void edac_pci_dev_parity_iterator(pci_parity_check_fn_t fn)
795 struct pci_dev *dev = NULL;
797 /* request for kernel access to the next PCI device, if any,
798 * and while we are looking at it have its reference count
799 * bumped until we are done with it
801 while((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
802 /* if whitelist exists then it has priority, so only scan
803 * those devices on the whitelist
805 if (pci_whitelist_count > 0 ) {
806 if (check_dev_on_list(pci_whitelist,
807 pci_whitelist_count, dev))
811 * if no whitelist, then check if this devices is
814 if (!check_dev_on_list(pci_blacklist,
815 pci_blacklist_count, dev))
821 static void do_pci_parity_check(void)
826 debugf3("%s()\n", __func__);
828 if (!check_pci_parity)
831 before_count = atomic_read(&pci_parity_count);
833 /* scan all PCI devices looking for a Parity Error on devices and
836 local_irq_save(flags);
837 edac_pci_dev_parity_iterator(edac_pci_dev_parity_test);
838 local_irq_restore(flags);
840 /* Only if operator has selected panic on PCI Error */
841 if (panic_on_pci_parity) {
842 /* If the count is different 'after' from 'before' */
843 if (before_count != atomic_read(&pci_parity_count))
844 panic("EDAC: PCI Parity Error");
848 static inline void clear_pci_parity_errors(void)
850 /* Clear any PCI bus parity errors that devices initially have logged
851 * in their registers.
853 edac_pci_dev_parity_iterator(edac_pci_dev_parity_clear);
856 #else /* CONFIG_PCI */
858 static inline void do_pci_parity_check(void)
863 static inline void clear_pci_parity_errors(void)
868 static void edac_sysfs_pci_teardown(void)
872 static int edac_sysfs_pci_setup(void)
876 #endif /* CONFIG_PCI */
878 #ifndef DISABLE_EDAC_SYSFS
880 /* EDAC sysfs CSROW data structures and methods */
882 /* Set of more detailed csrow<id> attribute show/store functions */
883 static ssize_t csrow_ch0_dimm_label_show(struct csrow_info *csrow, char *data)
887 if (csrow->nr_channels > 0) {
888 size = snprintf(data, EDAC_MC_LABEL_LEN,"%s\n",
889 csrow->channels[0].label);
895 static ssize_t csrow_ch1_dimm_label_show(struct csrow_info *csrow, char *data)
899 if (csrow->nr_channels > 0) {
900 size = snprintf(data, EDAC_MC_LABEL_LEN, "%s\n",
901 csrow->channels[1].label);
907 static ssize_t csrow_ch0_dimm_label_store(struct csrow_info *csrow,
908 const char *data, size_t size)
910 ssize_t max_size = 0;
912 if (csrow->nr_channels > 0) {
913 max_size = min((ssize_t)size,(ssize_t)EDAC_MC_LABEL_LEN-1);
914 strncpy(csrow->channels[0].label, data, max_size);
915 csrow->channels[0].label[max_size] = '\0';
921 static ssize_t csrow_ch1_dimm_label_store(struct csrow_info *csrow,
922 const char *data, size_t size)
924 ssize_t max_size = 0;
926 if (csrow->nr_channels > 1) {
927 max_size = min((ssize_t)size,(ssize_t)EDAC_MC_LABEL_LEN-1);
928 strncpy(csrow->channels[1].label, data, max_size);
929 csrow->channels[1].label[max_size] = '\0';
935 static ssize_t csrow_ue_count_show(struct csrow_info *csrow, char *data)
937 return sprintf(data,"%u\n", csrow->ue_count);
940 static ssize_t csrow_ce_count_show(struct csrow_info *csrow, char *data)
942 return sprintf(data,"%u\n", csrow->ce_count);
945 static ssize_t csrow_ch0_ce_count_show(struct csrow_info *csrow, char *data)
949 if (csrow->nr_channels > 0) {
950 size = sprintf(data,"%u\n", csrow->channels[0].ce_count);
956 static ssize_t csrow_ch1_ce_count_show(struct csrow_info *csrow, char *data)
960 if (csrow->nr_channels > 1) {
961 size = sprintf(data,"%u\n", csrow->channels[1].ce_count);
967 static ssize_t csrow_size_show(struct csrow_info *csrow, char *data)
969 return sprintf(data,"%u\n", PAGES_TO_MiB(csrow->nr_pages));
972 static ssize_t csrow_mem_type_show(struct csrow_info *csrow, char *data)
974 return sprintf(data,"%s\n", mem_types[csrow->mtype]);
977 static ssize_t csrow_dev_type_show(struct csrow_info *csrow, char *data)
979 return sprintf(data,"%s\n", dev_types[csrow->dtype]);
982 static ssize_t csrow_edac_mode_show(struct csrow_info *csrow, char *data)
984 return sprintf(data,"%s\n", edac_caps[csrow->edac_mode]);
987 struct csrowdev_attribute {
988 struct attribute attr;
989 ssize_t (*show)(struct csrow_info *,char *);
990 ssize_t (*store)(struct csrow_info *, const char *,size_t);
993 #define to_csrow(k) container_of(k, struct csrow_info, kobj)
994 #define to_csrowdev_attr(a) container_of(a, struct csrowdev_attribute, attr)
996 /* Set of show/store higher level functions for csrow objects */
997 static ssize_t csrowdev_show(struct kobject *kobj, struct attribute *attr,
1000 struct csrow_info *csrow = to_csrow(kobj);
1001 struct csrowdev_attribute *csrowdev_attr = to_csrowdev_attr(attr);
1003 if (csrowdev_attr->show)
1004 return csrowdev_attr->show(csrow, buffer);
1009 static ssize_t csrowdev_store(struct kobject *kobj, struct attribute *attr,
1010 const char *buffer, size_t count)
1012 struct csrow_info *csrow = to_csrow(kobj);
1013 struct csrowdev_attribute * csrowdev_attr = to_csrowdev_attr(attr);
1015 if (csrowdev_attr->store)
1016 return csrowdev_attr->store(csrow, buffer, count);
1021 static struct sysfs_ops csrowfs_ops = {
1022 .show = csrowdev_show,
1023 .store = csrowdev_store
1026 #define CSROWDEV_ATTR(_name,_mode,_show,_store) \
1027 struct csrowdev_attribute attr_##_name = { \
1028 .attr = {.name = __stringify(_name), .mode = _mode }, \
1033 /* cwrow<id>/attribute files */
1034 CSROWDEV_ATTR(size_mb,S_IRUGO,csrow_size_show,NULL);
1035 CSROWDEV_ATTR(dev_type,S_IRUGO,csrow_dev_type_show,NULL);
1036 CSROWDEV_ATTR(mem_type,S_IRUGO,csrow_mem_type_show,NULL);
1037 CSROWDEV_ATTR(edac_mode,S_IRUGO,csrow_edac_mode_show,NULL);
1038 CSROWDEV_ATTR(ue_count,S_IRUGO,csrow_ue_count_show,NULL);
1039 CSROWDEV_ATTR(ce_count,S_IRUGO,csrow_ce_count_show,NULL);
1040 CSROWDEV_ATTR(ch0_ce_count,S_IRUGO,csrow_ch0_ce_count_show,NULL);
1041 CSROWDEV_ATTR(ch1_ce_count,S_IRUGO,csrow_ch1_ce_count_show,NULL);
1043 /* control/attribute files */
1044 CSROWDEV_ATTR(ch0_dimm_label,S_IRUGO|S_IWUSR,
1045 csrow_ch0_dimm_label_show,
1046 csrow_ch0_dimm_label_store);
1047 CSROWDEV_ATTR(ch1_dimm_label,S_IRUGO|S_IWUSR,
1048 csrow_ch1_dimm_label_show,
1049 csrow_ch1_dimm_label_store);
1051 /* Attributes of the CSROW<id> object */
1052 static struct csrowdev_attribute *csrow_attr[] = {
1061 &attr_ch0_dimm_label,
1062 &attr_ch1_dimm_label,
1066 /* No memory to release */
1067 static void edac_csrow_instance_release(struct kobject *kobj)
1069 struct csrow_info *cs;
1071 debugf1("%s()\n", __func__);
1072 cs = container_of(kobj, struct csrow_info, kobj);
1073 complete(&cs->kobj_complete);
1076 static struct kobj_type ktype_csrow = {
1077 .release = edac_csrow_instance_release,
1078 .sysfs_ops = &csrowfs_ops,
1079 .default_attrs = (struct attribute **) csrow_attr,
1082 /* Create a CSROW object under specifed edac_mc_device */
1083 static int edac_create_csrow_object(struct kobject *edac_mci_kobj,
1084 struct csrow_info *csrow, int index)
1088 debugf0("%s()\n", __func__);
1089 memset(&csrow->kobj, 0, sizeof(csrow->kobj));
1091 /* generate ..../edac/mc/mc<id>/csrow<index> */
1093 csrow->kobj.parent = edac_mci_kobj;
1094 csrow->kobj.ktype = &ktype_csrow;
1096 /* name this instance of csrow<id> */
1097 err = kobject_set_name(&csrow->kobj,"csrow%d",index);
1100 /* Instanstiate the csrow object */
1101 err = kobject_register(&csrow->kobj);
1104 debugf0("Failed to register CSROW%d\n",index);
1106 debugf0("Registered CSROW%d\n",index);
1112 /* sysfs data structures and methods for the MCI kobjects */
1114 static ssize_t mci_reset_counters_store(struct mem_ctl_info *mci,
1115 const char *data, size_t count)
1119 mci->ue_noinfo_count = 0;
1120 mci->ce_noinfo_count = 0;
1124 for (row = 0; row < mci->nr_csrows; row++) {
1125 struct csrow_info *ri = &mci->csrows[row];
1130 for (chan = 0; chan < ri->nr_channels; chan++)
1131 ri->channels[chan].ce_count = 0;
1134 mci->start_time = jiffies;
1138 static ssize_t mci_ue_count_show(struct mem_ctl_info *mci, char *data)
1140 return sprintf(data,"%d\n", mci->ue_count);
1143 static ssize_t mci_ce_count_show(struct mem_ctl_info *mci, char *data)
1145 return sprintf(data,"%d\n", mci->ce_count);
1148 static ssize_t mci_ce_noinfo_show(struct mem_ctl_info *mci, char *data)
1150 return sprintf(data,"%d\n", mci->ce_noinfo_count);
1153 static ssize_t mci_ue_noinfo_show(struct mem_ctl_info *mci, char *data)
1155 return sprintf(data,"%d\n", mci->ue_noinfo_count);
1158 static ssize_t mci_seconds_show(struct mem_ctl_info *mci, char *data)
1160 return sprintf(data,"%ld\n", (jiffies - mci->start_time) / HZ);
1163 static ssize_t mci_mod_name_show(struct mem_ctl_info *mci, char *data)
1165 return sprintf(data,"%s %s\n", mci->mod_name, mci->mod_ver);
1168 static ssize_t mci_ctl_name_show(struct mem_ctl_info *mci, char *data)
1170 return sprintf(data,"%s\n", mci->ctl_name);
1173 static int mci_output_edac_cap(char *buf, unsigned long edac_cap)
1178 for (bit_idx = 0; bit_idx < 8 * sizeof(edac_cap); bit_idx++) {
1179 if ((edac_cap >> bit_idx) & 0x1)
1180 p += sprintf(p, "%s ", edac_caps[bit_idx]);
1186 static ssize_t mci_edac_capability_show(struct mem_ctl_info *mci, char *data)
1190 p += mci_output_edac_cap(p,mci->edac_ctl_cap);
1191 p += sprintf(p, "\n");
1195 static ssize_t mci_edac_current_capability_show(struct mem_ctl_info *mci,
1200 p += mci_output_edac_cap(p,mci->edac_cap);
1201 p += sprintf(p, "\n");
1205 static int mci_output_mtype_cap(char *buf, unsigned long mtype_cap)
1210 for (bit_idx = 0; bit_idx < 8 * sizeof(mtype_cap); bit_idx++) {
1211 if ((mtype_cap >> bit_idx) & 0x1)
1212 p += sprintf(p, "%s ", mem_types[bit_idx]);
1218 static ssize_t mci_supported_mem_type_show(struct mem_ctl_info *mci,
1223 p += mci_output_mtype_cap(p,mci->mtype_cap);
1224 p += sprintf(p, "\n");
1228 static ssize_t mci_size_mb_show(struct mem_ctl_info *mci, char *data)
1230 int total_pages, csrow_idx;
1232 for (total_pages = csrow_idx = 0; csrow_idx < mci->nr_csrows;
1234 struct csrow_info *csrow = &mci->csrows[csrow_idx];
1236 if (!csrow->nr_pages)
1239 total_pages += csrow->nr_pages;
1242 return sprintf(data,"%u\n", PAGES_TO_MiB(total_pages));
1245 struct mcidev_attribute {
1246 struct attribute attr;
1247 ssize_t (*show)(struct mem_ctl_info *,char *);
1248 ssize_t (*store)(struct mem_ctl_info *, const char *,size_t);
1251 #define to_mci(k) container_of(k, struct mem_ctl_info, edac_mci_kobj)
1252 #define to_mcidev_attr(a) container_of(a, struct mcidev_attribute, attr)
1254 static ssize_t mcidev_show(struct kobject *kobj, struct attribute *attr,
1257 struct mem_ctl_info *mem_ctl_info = to_mci(kobj);
1258 struct mcidev_attribute * mcidev_attr = to_mcidev_attr(attr);
1260 if (mcidev_attr->show)
1261 return mcidev_attr->show(mem_ctl_info, buffer);
1266 static ssize_t mcidev_store(struct kobject *kobj, struct attribute *attr,
1267 const char *buffer, size_t count)
1269 struct mem_ctl_info *mem_ctl_info = to_mci(kobj);
1270 struct mcidev_attribute * mcidev_attr = to_mcidev_attr(attr);
1272 if (mcidev_attr->store)
1273 return mcidev_attr->store(mem_ctl_info, buffer, count);
1278 static struct sysfs_ops mci_ops = {
1279 .show = mcidev_show,
1280 .store = mcidev_store
1283 #define MCIDEV_ATTR(_name,_mode,_show,_store) \
1284 struct mcidev_attribute mci_attr_##_name = { \
1285 .attr = {.name = __stringify(_name), .mode = _mode }, \
1291 MCIDEV_ATTR(reset_counters,S_IWUSR,NULL,mci_reset_counters_store);
1293 /* Attribute files */
1294 MCIDEV_ATTR(mc_name,S_IRUGO,mci_ctl_name_show,NULL);
1295 MCIDEV_ATTR(module_name,S_IRUGO,mci_mod_name_show,NULL);
1296 MCIDEV_ATTR(edac_capability,S_IRUGO,mci_edac_capability_show,NULL);
1297 MCIDEV_ATTR(size_mb,S_IRUGO,mci_size_mb_show,NULL);
1298 MCIDEV_ATTR(seconds_since_reset,S_IRUGO,mci_seconds_show,NULL);
1299 MCIDEV_ATTR(ue_noinfo_count,S_IRUGO,mci_ue_noinfo_show,NULL);
1300 MCIDEV_ATTR(ce_noinfo_count,S_IRUGO,mci_ce_noinfo_show,NULL);
1301 MCIDEV_ATTR(ue_count,S_IRUGO,mci_ue_count_show,NULL);
1302 MCIDEV_ATTR(ce_count,S_IRUGO,mci_ce_count_show,NULL);
1303 MCIDEV_ATTR(edac_current_capability,S_IRUGO,
1304 mci_edac_current_capability_show,NULL);
1305 MCIDEV_ATTR(supported_mem_type,S_IRUGO,
1306 mci_supported_mem_type_show,NULL);
1308 static struct mcidev_attribute *mci_attr[] = {
1309 &mci_attr_reset_counters,
1310 &mci_attr_module_name,
1312 &mci_attr_edac_capability,
1313 &mci_attr_edac_current_capability,
1314 &mci_attr_supported_mem_type,
1316 &mci_attr_seconds_since_reset,
1317 &mci_attr_ue_noinfo_count,
1318 &mci_attr_ce_noinfo_count,
1325 * Release of a MC controlling instance
1327 static void edac_mci_instance_release(struct kobject *kobj)
1329 struct mem_ctl_info *mci;
1332 debugf0("%s() idx=%d\n", __func__, mci->mc_idx);
1333 complete(&mci->kobj_complete);
1336 static struct kobj_type ktype_mci = {
1337 .release = edac_mci_instance_release,
1338 .sysfs_ops = &mci_ops,
1339 .default_attrs = (struct attribute **) mci_attr,
1342 #endif /* DISABLE_EDAC_SYSFS */
1344 #define EDAC_DEVICE_SYMLINK "device"
1347 * Create a new Memory Controller kobject instance,
1348 * mc<id> under the 'mc' directory
1354 static int edac_create_sysfs_mci_device(struct mem_ctl_info *mci)
1355 #ifdef DISABLE_EDAC_SYSFS
1363 struct csrow_info *csrow;
1364 struct kobject *edac_mci_kobj=&mci->edac_mci_kobj;
1366 debugf0("%s() idx=%d\n", __func__, mci->mc_idx);
1367 memset(edac_mci_kobj, 0, sizeof(*edac_mci_kobj));
1369 /* set the name of the mc<id> object */
1370 err = kobject_set_name(edac_mci_kobj,"mc%d",mci->mc_idx);
1375 /* link to our parent the '..../edac/mc' object */
1376 edac_mci_kobj->parent = &edac_memctrl_kobj;
1377 edac_mci_kobj->ktype = &ktype_mci;
1379 /* register the mc<id> kobject */
1380 err = kobject_register(edac_mci_kobj);
1385 /* create a symlink for the device */
1386 err = sysfs_create_link(edac_mci_kobj, &mci->dev->kobj,
1387 EDAC_DEVICE_SYMLINK);
1392 /* Make directories for each CSROW object
1393 * under the mc<id> kobject
1395 for (i = 0; i < mci->nr_csrows; i++) {
1396 csrow = &mci->csrows[i];
1398 /* Only expose populated CSROWs */
1399 if (csrow->nr_pages > 0) {
1400 err = edac_create_csrow_object(edac_mci_kobj,csrow,i);
1409 /* CSROW error: backout what has already been registered, */
1411 for ( i--; i >= 0; i--) {
1412 if (csrow->nr_pages > 0) {
1413 init_completion(&csrow->kobj_complete);
1414 kobject_unregister(&mci->csrows[i].kobj);
1415 wait_for_completion(&csrow->kobj_complete);
1420 init_completion(&mci->kobj_complete);
1421 kobject_unregister(edac_mci_kobj);
1422 wait_for_completion(&mci->kobj_complete);
1425 #endif /* DISABLE_EDAC_SYSFS */
1428 * remove a Memory Controller instance
1430 static void edac_remove_sysfs_mci_device(struct mem_ctl_info *mci)
1432 #ifndef DISABLE_EDAC_SYSFS
1435 debugf0("%s()\n", __func__);
1437 /* remove all csrow kobjects */
1438 for (i = 0; i < mci->nr_csrows; i++) {
1439 if (mci->csrows[i].nr_pages > 0) {
1440 init_completion(&mci->csrows[i].kobj_complete);
1441 kobject_unregister(&mci->csrows[i].kobj);
1442 wait_for_completion(&mci->csrows[i].kobj_complete);
1446 sysfs_remove_link(&mci->edac_mci_kobj, EDAC_DEVICE_SYMLINK);
1447 init_completion(&mci->kobj_complete);
1448 kobject_unregister(&mci->edac_mci_kobj);
1449 wait_for_completion(&mci->kobj_complete);
1450 #endif /* DISABLE_EDAC_SYSFS */
1453 /* END OF sysfs data and methods */
1455 #ifdef CONFIG_EDAC_DEBUG
1457 void edac_mc_dump_channel(struct channel_info *chan)
1459 debugf4("\tchannel = %p\n", chan);
1460 debugf4("\tchannel->chan_idx = %d\n", chan->chan_idx);
1461 debugf4("\tchannel->ce_count = %d\n", chan->ce_count);
1462 debugf4("\tchannel->label = '%s'\n", chan->label);
1463 debugf4("\tchannel->csrow = %p\n\n", chan->csrow);
1465 EXPORT_SYMBOL_GPL(edac_mc_dump_channel);
1467 void edac_mc_dump_csrow(struct csrow_info *csrow)
1469 debugf4("\tcsrow = %p\n", csrow);
1470 debugf4("\tcsrow->csrow_idx = %d\n", csrow->csrow_idx);
1471 debugf4("\tcsrow->first_page = 0x%lx\n",
1473 debugf4("\tcsrow->last_page = 0x%lx\n", csrow->last_page);
1474 debugf4("\tcsrow->page_mask = 0x%lx\n", csrow->page_mask);
1475 debugf4("\tcsrow->nr_pages = 0x%x\n", csrow->nr_pages);
1476 debugf4("\tcsrow->nr_channels = %d\n",
1477 csrow->nr_channels);
1478 debugf4("\tcsrow->channels = %p\n", csrow->channels);
1479 debugf4("\tcsrow->mci = %p\n\n", csrow->mci);
1481 EXPORT_SYMBOL_GPL(edac_mc_dump_csrow);
1483 void edac_mc_dump_mci(struct mem_ctl_info *mci)
1485 debugf3("\tmci = %p\n", mci);
1486 debugf3("\tmci->mtype_cap = %lx\n", mci->mtype_cap);
1487 debugf3("\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap);
1488 debugf3("\tmci->edac_cap = %lx\n", mci->edac_cap);
1489 debugf4("\tmci->edac_check = %p\n", mci->edac_check);
1490 debugf3("\tmci->nr_csrows = %d, csrows = %p\n",
1491 mci->nr_csrows, mci->csrows);
1492 debugf3("\tdev = %p\n", mci->dev);
1493 debugf3("\tmod_name:ctl_name = %s:%s\n",
1494 mci->mod_name, mci->ctl_name);
1495 debugf3("\tpvt_info = %p\n\n", mci->pvt_info);
1497 EXPORT_SYMBOL_GPL(edac_mc_dump_mci);
1499 #endif /* CONFIG_EDAC_DEBUG */
1501 /* 'ptr' points to a possibly unaligned item X such that sizeof(X) is 'size'.
1502 * Adjust 'ptr' so that its alignment is at least as stringent as what the
1503 * compiler would provide for X and return the aligned result.
1505 * If 'size' is a constant, the compiler will optimize this whole function
1506 * down to either a no-op or the addition of a constant to the value of 'ptr'.
1508 static inline char * align_ptr(void *ptr, unsigned size)
1512 /* Here we assume that the alignment of a "long long" is the most
1513 * stringent alignment that the compiler will ever provide by default.
1514 * As far as I know, this is a reasonable assumption.
1516 if (size > sizeof(long))
1517 align = sizeof(long long);
1518 else if (size > sizeof(int))
1519 align = sizeof(long);
1520 else if (size > sizeof(short))
1521 align = sizeof(int);
1522 else if (size > sizeof(char))
1523 align = sizeof(short);
1525 return (char *) ptr;
1530 return (char *) ptr;
1532 return (char *) (((unsigned long) ptr) + align - r);
1536 * edac_mc_alloc: Allocate a struct mem_ctl_info structure
1537 * @size_pvt: size of private storage needed
1538 * @nr_csrows: Number of CWROWS needed for this MC
1539 * @nr_chans: Number of channels for the MC
1541 * Everything is kmalloc'ed as one big chunk - more efficient.
1542 * Only can be used if all structures have the same lifetime - otherwise
1543 * you have to allocate and initialize your own structures.
1545 * Use edac_mc_free() to free mc structures allocated by this function.
1548 * NULL allocation failed
1549 * struct mem_ctl_info pointer
1551 struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows,
1554 struct mem_ctl_info *mci;
1555 struct csrow_info *csi, *csrow;
1556 struct channel_info *chi, *chp, *chan;
1561 /* Figure out the offsets of the various items from the start of an mc
1562 * structure. We want the alignment of each item to be at least as
1563 * stringent as what the compiler would provide if we could simply
1564 * hardcode everything into a single struct.
1566 mci = (struct mem_ctl_info *) 0;
1567 csi = (struct csrow_info *)align_ptr(&mci[1], sizeof(*csi));
1568 chi = (struct channel_info *)
1569 align_ptr(&csi[nr_csrows], sizeof(*chi));
1570 pvt = align_ptr(&chi[nr_chans * nr_csrows], sz_pvt);
1571 size = ((unsigned long) pvt) + sz_pvt;
1573 if ((mci = kmalloc(size, GFP_KERNEL)) == NULL)
1576 /* Adjust pointers so they point within the memory we just allocated
1577 * rather than an imaginary chunk of memory located at address 0.
1579 csi = (struct csrow_info *) (((char *) mci) + ((unsigned long) csi));
1580 chi = (struct channel_info *) (((char *) mci) + ((unsigned long) chi));
1581 pvt = sz_pvt ? (((char *) mci) + ((unsigned long) pvt)) : NULL;
1583 memset(mci, 0, size); /* clear all fields */
1585 mci->pvt_info = pvt;
1586 mci->nr_csrows = nr_csrows;
1588 for (row = 0; row < nr_csrows; row++) {
1590 csrow->csrow_idx = row;
1592 csrow->nr_channels = nr_chans;
1593 chp = &chi[row * nr_chans];
1594 csrow->channels = chp;
1596 for (chn = 0; chn < nr_chans; chn++) {
1598 chan->chan_idx = chn;
1599 chan->csrow = csrow;
1605 EXPORT_SYMBOL_GPL(edac_mc_alloc);
1608 * edac_mc_free: Free a previously allocated 'mci' structure
1609 * @mci: pointer to a struct mem_ctl_info structure
1611 void edac_mc_free(struct mem_ctl_info *mci)
1615 EXPORT_SYMBOL_GPL(edac_mc_free);
1617 static struct mem_ctl_info *find_mci_by_dev(struct device *dev)
1619 struct mem_ctl_info *mci;
1620 struct list_head *item;
1622 debugf3("%s()\n", __func__);
1624 list_for_each(item, &mc_devices) {
1625 mci = list_entry(item, struct mem_ctl_info, link);
1627 if (mci->dev == dev)
1634 /* Return 0 on success, 1 on failure.
1635 * Before calling this function, caller must
1636 * assign a unique value to mci->mc_idx.
1638 static int add_mc_to_global_list (struct mem_ctl_info *mci)
1640 struct list_head *item, *insert_before;
1641 struct mem_ctl_info *p;
1643 insert_before = &mc_devices;
1645 if (unlikely((p = find_mci_by_dev(mci->dev)) != NULL))
1648 list_for_each(item, &mc_devices) {
1649 p = list_entry(item, struct mem_ctl_info, link);
1651 if (p->mc_idx >= mci->mc_idx) {
1652 if (unlikely(p->mc_idx == mci->mc_idx))
1655 insert_before = item;
1660 list_add_tail_rcu(&mci->link, insert_before);
1664 edac_printk(KERN_WARNING, EDAC_MC,
1665 "%s (%s) %s %s already assigned %d\n", p->dev->bus_id,
1666 dev_name(p->dev), p->mod_name, p->ctl_name, p->mc_idx);
1670 edac_printk(KERN_WARNING, EDAC_MC,
1671 "bug in low-level driver: attempt to assign\n"
1672 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
1676 static void complete_mc_list_del(struct rcu_head *head)
1678 struct mem_ctl_info *mci;
1680 mci = container_of(head, struct mem_ctl_info, rcu);
1681 INIT_LIST_HEAD(&mci->link);
1682 complete(&mci->complete);
1685 static void del_mc_from_global_list(struct mem_ctl_info *mci)
1687 list_del_rcu(&mci->link);
1688 init_completion(&mci->complete);
1689 call_rcu(&mci->rcu, complete_mc_list_del);
1690 wait_for_completion(&mci->complete);
1694 * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and
1695 * create sysfs entries associated with mci structure
1696 * @mci: pointer to the mci structure to be added to the list
1697 * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure.
1704 /* FIXME - should a warning be printed if no error detection? correction? */
1705 int edac_mc_add_mc(struct mem_ctl_info *mci, int mc_idx)
1707 debugf0("%s()\n", __func__);
1708 mci->mc_idx = mc_idx;
1709 #ifdef CONFIG_EDAC_DEBUG
1710 if (edac_debug_level >= 3)
1711 edac_mc_dump_mci(mci);
1713 if (edac_debug_level >= 4) {
1716 for (i = 0; i < mci->nr_csrows; i++) {
1719 edac_mc_dump_csrow(&mci->csrows[i]);
1720 for (j = 0; j < mci->csrows[i].nr_channels; j++)
1721 edac_mc_dump_channel(
1722 &mci->csrows[i].channels[j]);
1726 down(&mem_ctls_mutex);
1728 if (add_mc_to_global_list(mci))
1731 /* set load time so that error rate can be tracked */
1732 mci->start_time = jiffies;
1734 if (edac_create_sysfs_mci_device(mci)) {
1735 edac_mc_printk(mci, KERN_WARNING,
1736 "failed to create sysfs device\n");
1740 /* Report action taken */
1741 edac_mc_printk(mci, KERN_INFO, "Giving out device to %s %s: DEV %s\n",
1742 mci->mod_name, mci->ctl_name, dev_name(mci->dev));
1744 up(&mem_ctls_mutex);
1748 del_mc_from_global_list(mci);
1751 up(&mem_ctls_mutex);
1754 EXPORT_SYMBOL_GPL(edac_mc_add_mc);
1757 * edac_mc_del_mc: Remove sysfs entries for specified mci structure and
1758 * remove mci structure from global list
1759 * @pdev: Pointer to 'struct device' representing mci structure to remove.
1761 * Return pointer to removed mci structure, or NULL if device not found.
1763 struct mem_ctl_info * edac_mc_del_mc(struct device *dev)
1765 struct mem_ctl_info *mci;
1767 debugf0("MC: %s()\n", __func__);
1768 down(&mem_ctls_mutex);
1770 if ((mci = find_mci_by_dev(dev)) == NULL) {
1771 up(&mem_ctls_mutex);
1775 edac_remove_sysfs_mci_device(mci);
1776 del_mc_from_global_list(mci);
1777 up(&mem_ctls_mutex);
1778 edac_printk(KERN_INFO, EDAC_MC,
1779 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
1780 mci->mod_name, mci->ctl_name, dev_name(mci->dev));
1783 EXPORT_SYMBOL_GPL(edac_mc_del_mc);
1785 void edac_mc_scrub_block(unsigned long page, unsigned long offset, u32 size)
1789 unsigned long flags = 0;
1791 debugf3("%s()\n", __func__);
1793 /* ECC error page was not in our memory. Ignore it. */
1794 if(!pfn_valid(page))
1797 /* Find the actual page structure then map it and fix */
1798 pg = pfn_to_page(page);
1800 if (PageHighMem(pg))
1801 local_irq_save(flags);
1803 virt_addr = kmap_atomic(pg, KM_BOUNCE_READ);
1805 /* Perform architecture specific atomic scrub operation */
1806 atomic_scrub(virt_addr + offset, size);
1808 /* Unmap and complete */
1809 kunmap_atomic(virt_addr, KM_BOUNCE_READ);
1811 if (PageHighMem(pg))
1812 local_irq_restore(flags);
1814 EXPORT_SYMBOL_GPL(edac_mc_scrub_block);
1816 /* FIXME - should return -1 */
1817 int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
1819 struct csrow_info *csrows = mci->csrows;
1822 debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page);
1825 for (i = 0; i < mci->nr_csrows; i++) {
1826 struct csrow_info *csrow = &csrows[i];
1828 if (csrow->nr_pages == 0)
1831 debugf3("MC%d: %s(): first(0x%lx) page(0x%lx) last(0x%lx) "
1832 "mask(0x%lx)\n", mci->mc_idx, __func__,
1833 csrow->first_page, page, csrow->last_page,
1836 if ((page >= csrow->first_page) &&
1837 (page <= csrow->last_page) &&
1838 ((page & csrow->page_mask) ==
1839 (csrow->first_page & csrow->page_mask))) {
1846 edac_mc_printk(mci, KERN_ERR,
1847 "could not look up page error address %lx\n",
1848 (unsigned long) page);
1852 EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
1854 /* FIXME - setable log (warning/emerg) levels */
1855 /* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */
1856 void edac_mc_handle_ce(struct mem_ctl_info *mci,
1857 unsigned long page_frame_number, unsigned long offset_in_page,
1858 unsigned long syndrome, int row, int channel, const char *msg)
1860 unsigned long remapped_page;
1862 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
1864 /* FIXME - maybe make panic on INTERNAL ERROR an option */
1865 if (row >= mci->nr_csrows || row < 0) {
1866 /* something is wrong */
1867 edac_mc_printk(mci, KERN_ERR,
1868 "INTERNAL ERROR: row out of range "
1869 "(%d >= %d)\n", row, mci->nr_csrows);
1870 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
1874 if (channel >= mci->csrows[row].nr_channels || channel < 0) {
1875 /* something is wrong */
1876 edac_mc_printk(mci, KERN_ERR,
1877 "INTERNAL ERROR: channel out of range "
1878 "(%d >= %d)\n", channel,
1879 mci->csrows[row].nr_channels);
1880 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
1885 /* FIXME - put in DIMM location */
1886 edac_mc_printk(mci, KERN_WARNING,
1887 "CE page 0x%lx, offset 0x%lx, grain %d, syndrome "
1888 "0x%lx, row %d, channel %d, label \"%s\": %s\n",
1889 page_frame_number, offset_in_page,
1890 mci->csrows[row].grain, syndrome, row, channel,
1891 mci->csrows[row].channels[channel].label, msg);
1894 mci->csrows[row].ce_count++;
1895 mci->csrows[row].channels[channel].ce_count++;
1897 if (mci->scrub_mode & SCRUB_SW_SRC) {
1899 * Some MC's can remap memory so that it is still available
1900 * at a different address when PCI devices map into memory.
1901 * MC's that can't do this lose the memory where PCI devices
1902 * are mapped. This mapping is MC dependant and so we call
1903 * back into the MC driver for it to map the MC page to
1904 * a physical (CPU) page which can then be mapped to a virtual
1905 * page - which can then be scrubbed.
1907 remapped_page = mci->ctl_page_to_phys ?
1908 mci->ctl_page_to_phys(mci, page_frame_number) :
1911 edac_mc_scrub_block(remapped_page, offset_in_page,
1912 mci->csrows[row].grain);
1915 EXPORT_SYMBOL_GPL(edac_mc_handle_ce);
1917 void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg)
1920 edac_mc_printk(mci, KERN_WARNING,
1921 "CE - no information available: %s\n", msg);
1923 mci->ce_noinfo_count++;
1926 EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info);
1928 void edac_mc_handle_ue(struct mem_ctl_info *mci,
1929 unsigned long page_frame_number, unsigned long offset_in_page,
1930 int row, const char *msg)
1932 int len = EDAC_MC_LABEL_LEN * 4;
1933 char labels[len + 1];
1938 debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
1940 /* FIXME - maybe make panic on INTERNAL ERROR an option */
1941 if (row >= mci->nr_csrows || row < 0) {
1942 /* something is wrong */
1943 edac_mc_printk(mci, KERN_ERR,
1944 "INTERNAL ERROR: row out of range "
1945 "(%d >= %d)\n", row, mci->nr_csrows);
1946 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
1950 chars = snprintf(pos, len + 1, "%s",
1951 mci->csrows[row].channels[0].label);
1955 for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0);
1957 chars = snprintf(pos, len + 1, ":%s",
1958 mci->csrows[row].channels[chan].label);
1964 edac_mc_printk(mci, KERN_EMERG,
1965 "UE page 0x%lx, offset 0x%lx, grain %d, row %d, "
1966 "labels \"%s\": %s\n", page_frame_number,
1967 offset_in_page, mci->csrows[row].grain, row, labels,
1971 panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, "
1972 "row %d, labels \"%s\": %s\n", mci->mc_idx,
1973 page_frame_number, offset_in_page,
1974 mci->csrows[row].grain, row, labels, msg);
1977 mci->csrows[row].ue_count++;
1979 EXPORT_SYMBOL_GPL(edac_mc_handle_ue);
1981 void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg)
1984 panic("EDAC MC%d: Uncorrected Error", mci->mc_idx);
1987 edac_mc_printk(mci, KERN_WARNING,
1988 "UE - no information available: %s\n", msg);
1989 mci->ue_noinfo_count++;
1992 EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info);
1996 * Iterate over all MC instances and check for ECC, et al, errors
1998 static inline void check_mc_devices(void)
2000 struct list_head *item;
2001 struct mem_ctl_info *mci;
2003 debugf3("%s()\n", __func__);
2004 down(&mem_ctls_mutex);
2006 list_for_each(item, &mc_devices) {
2007 mci = list_entry(item, struct mem_ctl_info, link);
2009 if (mci->edac_check != NULL)
2010 mci->edac_check(mci);
2013 up(&mem_ctls_mutex);
2017 * Check MC status every poll_msec.
2018 * Check PCI status every poll_msec as well.
2020 * This where the work gets done for edac.
2022 * SMP safe, doesn't use NMI, and auto-rate-limits.
2024 static void do_edac_check(void)
2026 debugf3("%s()\n", __func__);
2028 do_pci_parity_check();
2031 static int edac_kernel_thread(void *arg)
2033 while (!kthread_should_stop()) {
2036 /* goto sleep for the interval */
2037 schedule_timeout_interruptible((HZ * poll_msec) / 1000);
2046 * module initialization entry point
2048 static int __init edac_mc_init(void)
2050 edac_printk(KERN_INFO, EDAC_MC, EDAC_MC_VERSION "\n");
2053 * Harvest and clear any boot/initialization PCI parity errors
2055 * FIXME: This only clears errors logged by devices present at time of
2056 * module initialization. We should also do an initial clear
2057 * of each newly hotplugged device.
2059 clear_pci_parity_errors();
2061 /* Create the MC sysfs entries */
2062 if (edac_sysfs_memctrl_setup()) {
2063 edac_printk(KERN_ERR, EDAC_MC,
2064 "Error initializing sysfs code\n");
2068 /* Create the PCI parity sysfs entries */
2069 if (edac_sysfs_pci_setup()) {
2070 edac_sysfs_memctrl_teardown();
2071 edac_printk(KERN_ERR, EDAC_MC,
2072 "EDAC PCI: Error initializing sysfs code\n");
2076 /* create our kernel thread */
2077 edac_thread = kthread_run(edac_kernel_thread, NULL, "kedac");
2079 if (IS_ERR(edac_thread)) {
2080 /* remove the sysfs entries */
2081 edac_sysfs_memctrl_teardown();
2082 edac_sysfs_pci_teardown();
2083 return PTR_ERR(edac_thread);
2091 * module exit/termination functioni
2093 static void __exit edac_mc_exit(void)
2095 debugf0("%s()\n", __func__);
2096 kthread_stop(edac_thread);
2098 /* tear down the sysfs device */
2099 edac_sysfs_memctrl_teardown();
2100 edac_sysfs_pci_teardown();
2103 module_init(edac_mc_init);
2104 module_exit(edac_mc_exit);
2106 MODULE_LICENSE("GPL");
2107 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh et al\n"
2108 "Based on work by Dan Hollis et al");
2109 MODULE_DESCRIPTION("Core library routines for MC reporting");
2111 module_param(panic_on_ue, int, 0644);
2112 MODULE_PARM_DESC(panic_on_ue, "Panic on uncorrected error: 0=off 1=on");
2114 module_param(check_pci_parity, int, 0644);
2115 MODULE_PARM_DESC(check_pci_parity, "Check for PCI bus parity errors: 0=off 1=on");
2116 module_param(panic_on_pci_parity, int, 0644);
2117 MODULE_PARM_DESC(panic_on_pci_parity, "Panic on PCI Bus Parity error: 0=off 1=on");
2119 module_param(log_ue, int, 0644);
2120 MODULE_PARM_DESC(log_ue, "Log uncorrectable error to console: 0=off 1=on");
2121 module_param(log_ce, int, 0644);
2122 MODULE_PARM_DESC(log_ce, "Log correctable error to console: 0=off 1=on");
2123 module_param(poll_msec, int, 0644);
2124 MODULE_PARM_DESC(poll_msec, "Polling period in milliseconds");
2125 #ifdef CONFIG_EDAC_DEBUG
2126 module_param(edac_debug_level, int, 0644);
2127 MODULE_PARM_DESC(edac_debug_level, "Debug level");