2 * (C) 2005, 2006 Linux Networx (http://lnxi.com)
3 * This file may be distributed under the terms of the
4 * GNU General Public License.
6 * Written Doug Thompson <norsk5@xmission.com>
9 #include <linux/module.h>
10 #include <linux/edac.h>
11 #include <linux/slab.h>
12 #include <linux/ctype.h>
14 #include "edac_core.h"
15 #include "edac_module.h"
17 /* Turn off this whole feature if PCI is not configured */
20 #define EDAC_PCI_SYMLINK "device"
22 /* data variables exported via sysfs */
23 static int check_pci_errors; /* default NO check PCI parity */
24 static int edac_pci_panic_on_pe; /* default NO panic on PCI Parity */
25 static int edac_pci_log_pe = 1; /* log PCI parity errors */
26 static int edac_pci_log_npe = 1; /* log PCI non-parity error errors */
27 static int edac_pci_poll_msec = 1000; /* one second workq period */
29 static atomic_t pci_parity_count = ATOMIC_INIT(0);
30 static atomic_t pci_nonparity_count = ATOMIC_INIT(0);
32 static struct kobject *edac_pci_top_main_kobj;
33 static atomic_t edac_pci_sysfs_refcount = ATOMIC_INIT(0);
35 /* getter functions for the data variables */
36 int edac_pci_get_check_errors(void)
38 return check_pci_errors;
41 static int edac_pci_get_log_pe(void)
43 return edac_pci_log_pe;
46 static int edac_pci_get_log_npe(void)
48 return edac_pci_log_npe;
51 static int edac_pci_get_panic_on_pe(void)
53 return edac_pci_panic_on_pe;
56 int edac_pci_get_poll_msec(void)
58 return edac_pci_poll_msec;
61 /**************************** EDAC PCI sysfs instance *******************/
62 static ssize_t instance_pe_count_show(struct edac_pci_ctl_info *pci, char *data)
64 return sprintf(data, "%u\n", atomic_read(&pci->counters.pe_count));
67 static ssize_t instance_npe_count_show(struct edac_pci_ctl_info *pci,
70 return sprintf(data, "%u\n", atomic_read(&pci->counters.npe_count));
73 #define to_instance(k) container_of(k, struct edac_pci_ctl_info, kobj)
74 #define to_instance_attr(a) container_of(a, struct instance_attribute, attr)
76 /* DEVICE instance kobject release() function */
77 static void edac_pci_instance_release(struct kobject *kobj)
79 struct edac_pci_ctl_info *pci;
83 /* Form pointer to containing struct, the pci control struct */
84 pci = to_instance(kobj);
86 /* decrement reference count on top main kobj */
87 kobject_put(edac_pci_top_main_kobj);
89 kfree(pci); /* Free the control struct */
92 /* instance specific attribute structure */
93 struct instance_attribute {
94 struct attribute attr;
95 ssize_t(*show) (struct edac_pci_ctl_info *, char *);
96 ssize_t(*store) (struct edac_pci_ctl_info *, const char *, size_t);
99 /* Function to 'show' fields from the edac_pci 'instance' structure */
100 static ssize_t edac_pci_instance_show(struct kobject *kobj,
101 struct attribute *attr, char *buffer)
103 struct edac_pci_ctl_info *pci = to_instance(kobj);
104 struct instance_attribute *instance_attr = to_instance_attr(attr);
106 if (instance_attr->show)
107 return instance_attr->show(pci, buffer);
111 /* Function to 'store' fields into the edac_pci 'instance' structure */
112 static ssize_t edac_pci_instance_store(struct kobject *kobj,
113 struct attribute *attr,
114 const char *buffer, size_t count)
116 struct edac_pci_ctl_info *pci = to_instance(kobj);
117 struct instance_attribute *instance_attr = to_instance_attr(attr);
119 if (instance_attr->store)
120 return instance_attr->store(pci, buffer, count);
125 static const struct sysfs_ops pci_instance_ops = {
126 .show = edac_pci_instance_show,
127 .store = edac_pci_instance_store
130 #define INSTANCE_ATTR(_name, _mode, _show, _store) \
131 static struct instance_attribute attr_instance_##_name = { \
132 .attr = {.name = __stringify(_name), .mode = _mode }, \
137 INSTANCE_ATTR(pe_count, S_IRUGO, instance_pe_count_show, NULL);
138 INSTANCE_ATTR(npe_count, S_IRUGO, instance_npe_count_show, NULL);
140 /* pci instance attributes */
141 static struct instance_attribute *pci_instance_attr[] = {
142 &attr_instance_pe_count,
143 &attr_instance_npe_count,
147 /* the ktype for a pci instance */
148 static struct kobj_type ktype_pci_instance = {
149 .release = edac_pci_instance_release,
150 .sysfs_ops = &pci_instance_ops,
151 .default_attrs = (struct attribute **)pci_instance_attr,
155 * edac_pci_create_instance_kobj
157 * construct one EDAC PCI instance's kobject for use
159 static int edac_pci_create_instance_kobj(struct edac_pci_ctl_info *pci, int idx)
161 struct kobject *main_kobj;
166 /* First bump the ref count on the top main kobj, which will
167 * track the number of PCI instances we have, and thus nest
168 * properly on keeping the module loaded
170 main_kobj = kobject_get(edac_pci_top_main_kobj);
176 /* And now register this new kobject under the main kobj */
177 err = kobject_init_and_add(&pci->kobj, &ktype_pci_instance,
178 edac_pci_top_main_kobj, "pci%d", idx);
180 debugf2("failed to register instance pci%d\n",
182 kobject_put(edac_pci_top_main_kobj);
186 kobject_uevent(&pci->kobj, KOBJ_ADD);
187 debugf1("Register instance 'pci%d' kobject\n", idx);
191 /* Error unwind statck */
197 * edac_pci_unregister_sysfs_instance_kobj
199 * unregister the kobj for the EDAC PCI instance
201 static void edac_pci_unregister_sysfs_instance_kobj(
202 struct edac_pci_ctl_info *pci)
206 /* Unregister the instance kobject and allow its release
207 * function release the main reference count and then
210 kobject_put(&pci->kobj);
213 /***************************** EDAC PCI sysfs root **********************/
214 #define to_edacpci(k) container_of(k, struct edac_pci_ctl_info, kobj)
215 #define to_edacpci_attr(a) container_of(a, struct edac_pci_attr, attr)
217 /* simple show/store functions for attributes */
218 static ssize_t edac_pci_int_show(void *ptr, char *buffer)
221 return sprintf(buffer, "%d\n", *value);
224 static ssize_t edac_pci_int_store(void *ptr, const char *buffer, size_t count)
228 if (isdigit(*buffer))
229 *value = simple_strtoul(buffer, NULL, 0);
234 struct edac_pci_dev_attribute {
235 struct attribute attr;
237 ssize_t(*show) (void *, char *);
238 ssize_t(*store) (void *, const char *, size_t);
241 /* Set of show/store abstract level functions for PCI Parity object */
242 static ssize_t edac_pci_dev_show(struct kobject *kobj, struct attribute *attr,
245 struct edac_pci_dev_attribute *edac_pci_dev;
246 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;
248 if (edac_pci_dev->show)
249 return edac_pci_dev->show(edac_pci_dev->value, buffer);
253 static ssize_t edac_pci_dev_store(struct kobject *kobj,
254 struct attribute *attr, const char *buffer,
257 struct edac_pci_dev_attribute *edac_pci_dev;
258 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;
260 if (edac_pci_dev->show)
261 return edac_pci_dev->store(edac_pci_dev->value, buffer, count);
265 static const struct sysfs_ops edac_pci_sysfs_ops = {
266 .show = edac_pci_dev_show,
267 .store = edac_pci_dev_store
270 #define EDAC_PCI_ATTR(_name,_mode,_show,_store) \
271 static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
272 .attr = {.name = __stringify(_name), .mode = _mode }, \
278 #define EDAC_PCI_STRING_ATTR(_name,_data,_mode,_show,_store) \
279 static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
280 .attr = {.name = __stringify(_name), .mode = _mode }, \
286 /* PCI Parity control files */
287 EDAC_PCI_ATTR(check_pci_errors, S_IRUGO | S_IWUSR, edac_pci_int_show,
289 EDAC_PCI_ATTR(edac_pci_log_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,
291 EDAC_PCI_ATTR(edac_pci_log_npe, S_IRUGO | S_IWUSR, edac_pci_int_show,
293 EDAC_PCI_ATTR(edac_pci_panic_on_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,
295 EDAC_PCI_ATTR(pci_parity_count, S_IRUGO, edac_pci_int_show, NULL);
296 EDAC_PCI_ATTR(pci_nonparity_count, S_IRUGO, edac_pci_int_show, NULL);
298 /* Base Attributes of the memory ECC object */
299 static struct edac_pci_dev_attribute *edac_pci_attr[] = {
300 &edac_pci_attr_check_pci_errors,
301 &edac_pci_attr_edac_pci_log_pe,
302 &edac_pci_attr_edac_pci_log_npe,
303 &edac_pci_attr_edac_pci_panic_on_pe,
304 &edac_pci_attr_pci_parity_count,
305 &edac_pci_attr_pci_nonparity_count,
310 * edac_pci_release_main_kobj
312 * This release function is called when the reference count to the
313 * passed kobj goes to zero.
315 * This kobj is the 'main' kobject that EDAC PCI instances
316 * link to, and thus provide for proper nesting counts
318 static void edac_pci_release_main_kobj(struct kobject *kobj)
320 debugf0("here to module_put(THIS_MODULE)\n");
324 /* last reference to top EDAC PCI kobject has been removed,
325 * NOW release our ref count on the core module
327 module_put(THIS_MODULE);
330 /* ktype struct for the EDAC PCI main kobj */
331 static struct kobj_type ktype_edac_pci_main_kobj = {
332 .release = edac_pci_release_main_kobj,
333 .sysfs_ops = &edac_pci_sysfs_ops,
334 .default_attrs = (struct attribute **)edac_pci_attr,
338 * edac_pci_main_kobj_setup()
340 * setup the sysfs for EDAC PCI attributes
341 * assumes edac_subsys has already been initialized
343 static int edac_pci_main_kobj_setup(void)
346 struct bus_type *edac_subsys;
350 /* check and count if we have already created the main kobject */
351 if (atomic_inc_return(&edac_pci_sysfs_refcount) != 1)
354 /* First time, so create the main kobject and its
355 * controls and attributes
357 edac_subsys = edac_get_sysfs_subsys();
358 if (edac_subsys == NULL) {
359 debugf1("no edac_subsys\n");
361 goto decrement_count_fail;
364 /* Bump the reference count on this module to ensure the
365 * modules isn't unloaded until we deconstruct the top
366 * level main kobj for EDAC PCI
368 if (!try_module_get(THIS_MODULE)) {
369 debugf1("try_module_get() failed\n");
374 edac_pci_top_main_kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
375 if (!edac_pci_top_main_kobj) {
376 debugf1("Failed to allocate\n");
381 /* Instanstiate the pci object */
382 err = kobject_init_and_add(edac_pci_top_main_kobj,
383 &ktype_edac_pci_main_kobj,
384 &edac_subsys->dev_root->kobj, "pci");
386 debugf1("Failed to register '.../edac/pci'\n");
387 goto kobject_init_and_add_fail;
390 /* At this point, to 'release' the top level kobject
391 * for EDAC PCI, then edac_pci_main_kobj_teardown()
392 * must be used, for resources to be cleaned up properly
394 kobject_uevent(edac_pci_top_main_kobj, KOBJ_ADD);
395 debugf1("Registered '.../edac/pci' kobject\n");
399 /* Error unwind statck */
400 kobject_init_and_add_fail:
401 kfree(edac_pci_top_main_kobj);
404 module_put(THIS_MODULE);
407 edac_put_sysfs_subsys();
409 decrement_count_fail:
410 /* if are on this error exit, nothing to tear down */
411 atomic_dec(&edac_pci_sysfs_refcount);
417 * edac_pci_main_kobj_teardown()
419 * if no longer linked (needed) remove the top level EDAC PCI
420 * kobject with its controls and attributes
422 static void edac_pci_main_kobj_teardown(void)
426 /* Decrement the count and only if no more controller instances
427 * are connected perform the unregisteration of the top level
430 if (atomic_dec_return(&edac_pci_sysfs_refcount) == 0) {
431 debugf0("called kobject_put on main kobj\n");
432 kobject_put(edac_pci_top_main_kobj);
434 edac_put_sysfs_subsys();
439 * edac_pci_create_sysfs
441 * Create the controls/attributes for the specified EDAC PCI device
443 int edac_pci_create_sysfs(struct edac_pci_ctl_info *pci)
446 struct kobject *edac_kobj = &pci->kobj;
448 debugf0("idx=%d\n", pci->pci_idx);
450 /* create the top main EDAC PCI kobject, IF needed */
451 err = edac_pci_main_kobj_setup();
455 /* Create this instance's kobject under the MAIN kobject */
456 err = edac_pci_create_instance_kobj(pci, pci->pci_idx);
458 goto unregister_cleanup;
460 err = sysfs_create_link(edac_kobj, &pci->dev->kobj, EDAC_PCI_SYMLINK);
462 debugf0("sysfs_create_link() returned err= %d\n",
469 /* Error unwind stack */
471 edac_pci_unregister_sysfs_instance_kobj(pci);
474 edac_pci_main_kobj_teardown();
480 * edac_pci_remove_sysfs
482 * remove the controls and attributes for this EDAC PCI device
484 void edac_pci_remove_sysfs(struct edac_pci_ctl_info *pci)
486 debugf0("index=%d\n", pci->pci_idx);
488 /* Remove the symlink */
489 sysfs_remove_link(&pci->kobj, EDAC_PCI_SYMLINK);
491 /* remove this PCI instance's sysfs entries */
492 edac_pci_unregister_sysfs_instance_kobj(pci);
494 /* Call the main unregister function, which will determine
495 * if this 'pci' is the last instance.
496 * If it is, the main kobject will be unregistered as a result
498 debugf0("calling edac_pci_main_kobj_teardown()\n");
499 edac_pci_main_kobj_teardown();
502 /************************ PCI error handling *************************/
503 static u16 get_pci_parity_status(struct pci_dev *dev, int secondary)
508 where = secondary ? PCI_SEC_STATUS : PCI_STATUS;
509 pci_read_config_word(dev, where, &status);
511 /* If we get back 0xFFFF then we must suspect that the card has been
512 * pulled but the Linux PCI layer has not yet finished cleaning up.
513 * We don't want to report on such devices
516 if (status == 0xFFFF) {
519 pci_read_config_dword(dev, 0, &sanity);
521 if (sanity == 0xFFFFFFFF)
525 status &= PCI_STATUS_DETECTED_PARITY | PCI_STATUS_SIG_SYSTEM_ERROR |
529 /* reset only the bits we are interested in */
530 pci_write_config_word(dev, where, status);
536 /* Clear any PCI parity errors logged by this device. */
537 static void edac_pci_dev_parity_clear(struct pci_dev *dev)
541 get_pci_parity_status(dev, 0);
543 /* read the device TYPE, looking for bridges */
544 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
546 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE)
547 get_pci_parity_status(dev, 1);
553 * Function to retrieve the current parity status
557 static void edac_pci_dev_parity_test(struct pci_dev *dev)
563 /* stop any interrupts until we can acquire the status */
564 local_irq_save(flags);
566 /* read the STATUS register on this device */
567 status = get_pci_parity_status(dev, 0);
569 /* read the device TYPE, looking for bridges */
570 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
572 local_irq_restore(flags);
574 debugf4("PCI STATUS= 0x%04x %s\n", status, dev_name(&dev->dev));
576 /* check the status reg for errors on boards NOT marked as broken
577 * if broken, we cannot trust any of the status bits
579 if (status && !dev->broken_parity_status) {
580 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {
581 edac_printk(KERN_CRIT, EDAC_PCI,
582 "Signaled System Error on %s\n",
584 atomic_inc(&pci_nonparity_count);
587 if (status & (PCI_STATUS_PARITY)) {
588 edac_printk(KERN_CRIT, EDAC_PCI,
589 "Master Data Parity Error on %s\n",
592 atomic_inc(&pci_parity_count);
595 if (status & (PCI_STATUS_DETECTED_PARITY)) {
596 edac_printk(KERN_CRIT, EDAC_PCI,
597 "Detected Parity Error on %s\n",
600 atomic_inc(&pci_parity_count);
605 debugf4("PCI HEADER TYPE= 0x%02x %s\n", header_type, dev_name(&dev->dev));
607 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) {
608 /* On bridges, need to examine secondary status register */
609 status = get_pci_parity_status(dev, 1);
611 debugf4("PCI SEC_STATUS= 0x%04x %s\n", status, dev_name(&dev->dev));
613 /* check the secondary status reg for errors,
614 * on NOT broken boards
616 if (status && !dev->broken_parity_status) {
617 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {
618 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
619 "Signaled System Error on %s\n",
621 atomic_inc(&pci_nonparity_count);
624 if (status & (PCI_STATUS_PARITY)) {
625 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
626 "Master Data Parity Error on "
627 "%s\n", pci_name(dev));
629 atomic_inc(&pci_parity_count);
632 if (status & (PCI_STATUS_DETECTED_PARITY)) {
633 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
634 "Detected Parity Error on %s\n",
637 atomic_inc(&pci_parity_count);
643 /* reduce some complexity in definition of the iterator */
644 typedef void (*pci_parity_check_fn_t) (struct pci_dev *dev);
647 * pci_dev parity list iterator
648 * Scan the PCI device list for one pass, looking for SERRORs
649 * Master Parity ERRORS or Parity ERRORs on primary or secondary devices
651 static inline void edac_pci_dev_parity_iterator(pci_parity_check_fn_t fn)
653 struct pci_dev *dev = NULL;
655 /* request for kernel access to the next PCI device, if any,
656 * and while we are looking at it have its reference count
657 * bumped until we are done with it
659 while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
665 * edac_pci_do_parity_check
667 * performs the actual PCI parity check operation
669 void edac_pci_do_parity_check(void)
675 /* if policy has PCI check off, leave now */
676 if (!check_pci_errors)
679 before_count = atomic_read(&pci_parity_count);
681 /* scan all PCI devices looking for a Parity Error on devices and
683 * The iterator calls pci_get_device() which might sleep, thus
684 * we cannot disable interrupts in this scan.
686 edac_pci_dev_parity_iterator(edac_pci_dev_parity_test);
688 /* Only if operator has selected panic on PCI Error */
689 if (edac_pci_get_panic_on_pe()) {
690 /* If the count is different 'after' from 'before' */
691 if (before_count != atomic_read(&pci_parity_count))
692 panic("EDAC: PCI Parity Error");
697 * edac_pci_clear_parity_errors
699 * function to perform an iteration over the PCI devices
700 * and clearn their current status
702 void edac_pci_clear_parity_errors(void)
704 /* Clear any PCI bus parity errors that devices initially have logged
705 * in their registers.
707 edac_pci_dev_parity_iterator(edac_pci_dev_parity_clear);
713 * Called to handle a PARITY ERROR event
715 void edac_pci_handle_pe(struct edac_pci_ctl_info *pci, const char *msg)
718 /* global PE counter incremented by edac_pci_do_parity_check() */
719 atomic_inc(&pci->counters.pe_count);
721 if (edac_pci_get_log_pe())
722 edac_pci_printk(pci, KERN_WARNING,
723 "Parity Error ctl: %s %d: %s\n",
724 pci->ctl_name, pci->pci_idx, msg);
727 * poke all PCI devices and see which one is the troublemaker
728 * panic() is called if set
730 edac_pci_do_parity_check();
732 EXPORT_SYMBOL_GPL(edac_pci_handle_pe);
736 * edac_pci_handle_npe
738 * Called to handle a NON-PARITY ERROR event
740 void edac_pci_handle_npe(struct edac_pci_ctl_info *pci, const char *msg)
743 /* global NPE counter incremented by edac_pci_do_parity_check() */
744 atomic_inc(&pci->counters.npe_count);
746 if (edac_pci_get_log_npe())
747 edac_pci_printk(pci, KERN_WARNING,
748 "Non-Parity Error ctl: %s %d: %s\n",
749 pci->ctl_name, pci->pci_idx, msg);
752 * poke all PCI devices and see which one is the troublemaker
753 * panic() is called if set
755 edac_pci_do_parity_check();
757 EXPORT_SYMBOL_GPL(edac_pci_handle_npe);
760 * Define the PCI parameter to the module
762 module_param(check_pci_errors, int, 0644);
763 MODULE_PARM_DESC(check_pci_errors,
764 "Check for PCI bus parity errors: 0=off 1=on");
765 module_param(edac_pci_panic_on_pe, int, 0644);
766 MODULE_PARM_DESC(edac_pci_panic_on_pe,
767 "Panic on PCI Bus Parity error: 0=off 1=on");
769 #endif /* CONFIG_PCI */