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/sysdev.h>
11 #include <linux/ctype.h>
13 #include "edac_core.h"
14 #include "edac_module.h"
16 /* Turn off this whole feature if PCI is not configured */
19 #define EDAC_PCI_SYMLINK "device"
21 /* data variables exported via sysfs */
22 static int check_pci_errors; /* default NO check PCI parity */
23 static int edac_pci_panic_on_pe; /* default NO panic on PCI Parity */
24 static int edac_pci_log_pe = 1; /* log PCI parity errors */
25 static int edac_pci_log_npe = 1; /* log PCI non-parity error errors */
26 static int edac_pci_poll_msec = 1000; /* one second workq period */
28 static atomic_t pci_parity_count = ATOMIC_INIT(0);
29 static atomic_t pci_nonparity_count = ATOMIC_INIT(0);
31 static struct kobject *edac_pci_top_main_kobj;
32 static atomic_t edac_pci_sysfs_refcount = ATOMIC_INIT(0);
34 /* getter functions for the data variables */
35 int edac_pci_get_check_errors(void)
37 return check_pci_errors;
40 static int edac_pci_get_log_pe(void)
42 return edac_pci_log_pe;
45 static int edac_pci_get_log_npe(void)
47 return edac_pci_log_npe;
50 static int edac_pci_get_panic_on_pe(void)
52 return edac_pci_panic_on_pe;
55 int edac_pci_get_poll_msec(void)
57 return edac_pci_poll_msec;
60 /**************************** EDAC PCI sysfs instance *******************/
61 static ssize_t instance_pe_count_show(struct edac_pci_ctl_info *pci, char *data)
63 return sprintf(data, "%u\n", atomic_read(&pci->counters.pe_count));
66 static ssize_t instance_npe_count_show(struct edac_pci_ctl_info *pci,
69 return sprintf(data, "%u\n", atomic_read(&pci->counters.npe_count));
72 #define to_instance(k) container_of(k, struct edac_pci_ctl_info, kobj)
73 #define to_instance_attr(a) container_of(a, struct instance_attribute, attr)
75 /* DEVICE instance kobject release() function */
76 static void edac_pci_instance_release(struct kobject *kobj)
78 struct edac_pci_ctl_info *pci;
80 debugf0("%s()\n", __func__);
82 /* Form pointer to containing struct, the pci control struct */
83 pci = to_instance(kobj);
85 /* decrement reference count on top main kobj */
86 kobject_put(edac_pci_top_main_kobj);
88 kfree(pci); /* Free the control struct */
91 /* instance specific attribute structure */
92 struct instance_attribute {
93 struct attribute attr;
94 ssize_t(*show) (struct edac_pci_ctl_info *, char *);
95 ssize_t(*store) (struct edac_pci_ctl_info *, const char *, size_t);
98 /* Function to 'show' fields from the edac_pci 'instance' structure */
99 static ssize_t edac_pci_instance_show(struct kobject *kobj,
100 struct attribute *attr, char *buffer)
102 struct edac_pci_ctl_info *pci = to_instance(kobj);
103 struct instance_attribute *instance_attr = to_instance_attr(attr);
105 if (instance_attr->show)
106 return instance_attr->show(pci, buffer);
110 /* Function to 'store' fields into the edac_pci 'instance' structure */
111 static ssize_t edac_pci_instance_store(struct kobject *kobj,
112 struct attribute *attr,
113 const char *buffer, size_t count)
115 struct edac_pci_ctl_info *pci = to_instance(kobj);
116 struct instance_attribute *instance_attr = to_instance_attr(attr);
118 if (instance_attr->store)
119 return instance_attr->store(pci, buffer, count);
124 static struct sysfs_ops pci_instance_ops = {
125 .show = edac_pci_instance_show,
126 .store = edac_pci_instance_store
129 #define INSTANCE_ATTR(_name, _mode, _show, _store) \
130 static struct instance_attribute attr_instance_##_name = { \
131 .attr = {.name = __stringify(_name), .mode = _mode }, \
136 INSTANCE_ATTR(pe_count, S_IRUGO, instance_pe_count_show, NULL);
137 INSTANCE_ATTR(npe_count, S_IRUGO, instance_npe_count_show, NULL);
139 /* pci instance attributes */
140 static struct instance_attribute *pci_instance_attr[] = {
141 &attr_instance_pe_count,
142 &attr_instance_npe_count,
146 /* the ktype for a pci instance */
147 static struct kobj_type ktype_pci_instance = {
148 .release = edac_pci_instance_release,
149 .sysfs_ops = &pci_instance_ops,
150 .default_attrs = (struct attribute **)pci_instance_attr,
154 * edac_pci_create_instance_kobj
156 * construct one EDAC PCI instance's kobject for use
158 static int edac_pci_create_instance_kobj(struct edac_pci_ctl_info *pci, int idx)
160 struct kobject *main_kobj;
163 debugf0("%s()\n", __func__);
165 /* First bump the ref count on the top main kobj, which will
166 * track the number of PCI instances we have, and thus nest
167 * properly on keeping the module loaded
169 main_kobj = kobject_get(edac_pci_top_main_kobj);
175 /* And now register this new kobject under the main kobj */
176 err = kobject_init_and_add(&pci->kobj, &ktype_pci_instance,
177 edac_pci_top_main_kobj, "pci%d", idx);
179 debugf2("%s() failed to register instance pci%d\n",
181 kobject_put(edac_pci_top_main_kobj);
185 kobject_uevent(&pci->kobj, KOBJ_ADD);
186 debugf1("%s() Register instance 'pci%d' kobject\n", __func__, idx);
190 /* Error unwind statck */
196 * edac_pci_unregister_sysfs_instance_kobj
198 * unregister the kobj for the EDAC PCI instance
200 static void edac_pci_unregister_sysfs_instance_kobj(
201 struct edac_pci_ctl_info *pci)
203 debugf0("%s()\n", __func__);
205 /* Unregister the instance kobject and allow its release
206 * function release the main reference count and then
209 kobject_put(&pci->kobj);
212 /***************************** EDAC PCI sysfs root **********************/
213 #define to_edacpci(k) container_of(k, struct edac_pci_ctl_info, kobj)
214 #define to_edacpci_attr(a) container_of(a, struct edac_pci_attr, attr)
216 /* simple show/store functions for attributes */
217 static ssize_t edac_pci_int_show(void *ptr, char *buffer)
220 return sprintf(buffer, "%d\n", *value);
223 static ssize_t edac_pci_int_store(void *ptr, const char *buffer, size_t count)
227 if (isdigit(*buffer))
228 *value = simple_strtoul(buffer, NULL, 0);
233 struct edac_pci_dev_attribute {
234 struct attribute attr;
236 ssize_t(*show) (void *, char *);
237 ssize_t(*store) (void *, const char *, size_t);
240 /* Set of show/store abstract level functions for PCI Parity object */
241 static ssize_t edac_pci_dev_show(struct kobject *kobj, struct attribute *attr,
244 struct edac_pci_dev_attribute *edac_pci_dev;
245 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;
247 if (edac_pci_dev->show)
248 return edac_pci_dev->show(edac_pci_dev->value, buffer);
252 static ssize_t edac_pci_dev_store(struct kobject *kobj,
253 struct attribute *attr, const char *buffer,
256 struct edac_pci_dev_attribute *edac_pci_dev;
257 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;
259 if (edac_pci_dev->show)
260 return edac_pci_dev->store(edac_pci_dev->value, buffer, count);
264 static struct sysfs_ops edac_pci_sysfs_ops = {
265 .show = edac_pci_dev_show,
266 .store = edac_pci_dev_store
269 #define EDAC_PCI_ATTR(_name,_mode,_show,_store) \
270 static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
271 .attr = {.name = __stringify(_name), .mode = _mode }, \
277 #define EDAC_PCI_STRING_ATTR(_name,_data,_mode,_show,_store) \
278 static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
279 .attr = {.name = __stringify(_name), .mode = _mode }, \
285 /* PCI Parity control files */
286 EDAC_PCI_ATTR(check_pci_errors, S_IRUGO | S_IWUSR, edac_pci_int_show,
288 EDAC_PCI_ATTR(edac_pci_log_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,
290 EDAC_PCI_ATTR(edac_pci_log_npe, S_IRUGO | S_IWUSR, edac_pci_int_show,
292 EDAC_PCI_ATTR(edac_pci_panic_on_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,
294 EDAC_PCI_ATTR(pci_parity_count, S_IRUGO, edac_pci_int_show, NULL);
295 EDAC_PCI_ATTR(pci_nonparity_count, S_IRUGO, edac_pci_int_show, NULL);
297 /* Base Attributes of the memory ECC object */
298 static struct edac_pci_dev_attribute *edac_pci_attr[] = {
299 &edac_pci_attr_check_pci_errors,
300 &edac_pci_attr_edac_pci_log_pe,
301 &edac_pci_attr_edac_pci_log_npe,
302 &edac_pci_attr_edac_pci_panic_on_pe,
303 &edac_pci_attr_pci_parity_count,
304 &edac_pci_attr_pci_nonparity_count,
309 * edac_pci_release_main_kobj
311 * This release function is called when the reference count to the
312 * passed kobj goes to zero.
314 * This kobj is the 'main' kobject that EDAC PCI instances
315 * link to, and thus provide for proper nesting counts
317 static void edac_pci_release_main_kobj(struct kobject *kobj)
319 debugf0("%s() here to module_put(THIS_MODULE)\n", __func__);
323 /* last reference to top EDAC PCI kobject has been removed,
324 * NOW release our ref count on the core module
326 module_put(THIS_MODULE);
329 /* ktype struct for the EDAC PCI main kobj */
330 static struct kobj_type ktype_edac_pci_main_kobj = {
331 .release = edac_pci_release_main_kobj,
332 .sysfs_ops = &edac_pci_sysfs_ops,
333 .default_attrs = (struct attribute **)edac_pci_attr,
337 * edac_pci_main_kobj_setup()
339 * setup the sysfs for EDAC PCI attributes
340 * assumes edac_class has already been initialized
342 static int edac_pci_main_kobj_setup(void)
345 struct sysdev_class *edac_class;
347 debugf0("%s()\n", __func__);
349 /* check and count if we have already created the main kobject */
350 if (atomic_inc_return(&edac_pci_sysfs_refcount) != 1)
353 /* First time, so create the main kobject and its
354 * controls and atributes
356 edac_class = edac_get_edac_class();
357 if (edac_class == NULL) {
358 debugf1("%s() no edac_class\n", __func__);
360 goto decrement_count_fail;
363 /* Bump the reference count on this module to ensure the
364 * modules isn't unloaded until we deconstruct the top
365 * level main kobj for EDAC PCI
367 if (!try_module_get(THIS_MODULE)) {
368 debugf1("%s() try_module_get() failed\n", __func__);
370 goto decrement_count_fail;
373 edac_pci_top_main_kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
374 if (!edac_pci_top_main_kobj) {
375 debugf1("Failed to allocate\n");
380 /* Instanstiate the pci object */
381 err = kobject_init_and_add(edac_pci_top_main_kobj,
382 &ktype_edac_pci_main_kobj,
383 &edac_class->kset.kobj, "pci");
385 debugf1("Failed to register '.../edac/pci'\n");
386 goto kobject_init_and_add_fail;
389 /* At this point, to 'release' the top level kobject
390 * for EDAC PCI, then edac_pci_main_kobj_teardown()
391 * must be used, for resources to be cleaned up properly
393 kobject_uevent(edac_pci_top_main_kobj, KOBJ_ADD);
394 debugf1("Registered '.../edac/pci' kobject\n");
398 /* Error unwind statck */
399 kobject_init_and_add_fail:
400 kfree(edac_pci_top_main_kobj);
403 module_put(THIS_MODULE);
405 decrement_count_fail:
406 /* if are on this error exit, nothing to tear down */
407 atomic_dec(&edac_pci_sysfs_refcount);
413 * edac_pci_main_kobj_teardown()
415 * if no longer linked (needed) remove the top level EDAC PCI
416 * kobject with its controls and attributes
418 static void edac_pci_main_kobj_teardown(void)
420 debugf0("%s()\n", __func__);
422 /* Decrement the count and only if no more controller instances
423 * are connected perform the unregisteration of the top level
426 if (atomic_dec_return(&edac_pci_sysfs_refcount) == 0) {
427 debugf0("%s() called kobject_put on main kobj\n",
429 kobject_put(edac_pci_top_main_kobj);
435 * edac_pci_create_sysfs
437 * Create the controls/attributes for the specified EDAC PCI device
439 int edac_pci_create_sysfs(struct edac_pci_ctl_info *pci)
442 struct kobject *edac_kobj = &pci->kobj;
444 debugf0("%s() idx=%d\n", __func__, pci->pci_idx);
446 /* create the top main EDAC PCI kobject, IF needed */
447 err = edac_pci_main_kobj_setup();
451 /* Create this instance's kobject under the MAIN kobject */
452 err = edac_pci_create_instance_kobj(pci, pci->pci_idx);
454 goto unregister_cleanup;
456 err = sysfs_create_link(edac_kobj, &pci->dev->kobj, EDAC_PCI_SYMLINK);
458 debugf0("%s() sysfs_create_link() returned err= %d\n",
465 /* Error unwind stack */
467 edac_pci_unregister_sysfs_instance_kobj(pci);
470 edac_pci_main_kobj_teardown();
476 * edac_pci_remove_sysfs
478 * remove the controls and attributes for this EDAC PCI device
480 void edac_pci_remove_sysfs(struct edac_pci_ctl_info *pci)
482 debugf0("%s() index=%d\n", __func__, pci->pci_idx);
484 /* Remove the symlink */
485 sysfs_remove_link(&pci->kobj, EDAC_PCI_SYMLINK);
487 /* remove this PCI instance's sysfs entries */
488 edac_pci_unregister_sysfs_instance_kobj(pci);
490 /* Call the main unregister function, which will determine
491 * if this 'pci' is the last instance.
492 * If it is, the main kobject will be unregistered as a result
494 debugf0("%s() calling edac_pci_main_kobj_teardown()\n", __func__);
495 edac_pci_main_kobj_teardown();
498 /************************ PCI error handling *************************/
499 static u16 get_pci_parity_status(struct pci_dev *dev, int secondary)
504 where = secondary ? PCI_SEC_STATUS : PCI_STATUS;
505 pci_read_config_word(dev, where, &status);
507 /* If we get back 0xFFFF then we must suspect that the card has been
508 * pulled but the Linux PCI layer has not yet finished cleaning up.
509 * We don't want to report on such devices
512 if (status == 0xFFFF) {
515 pci_read_config_dword(dev, 0, &sanity);
517 if (sanity == 0xFFFFFFFF)
521 status &= PCI_STATUS_DETECTED_PARITY | PCI_STATUS_SIG_SYSTEM_ERROR |
525 /* reset only the bits we are interested in */
526 pci_write_config_word(dev, where, status);
532 /* Clear any PCI parity errors logged by this device. */
533 static void edac_pci_dev_parity_clear(struct pci_dev *dev)
537 debugf0("%s()\n", __func__);
539 get_pci_parity_status(dev, 0);
541 /* read the device TYPE, looking for bridges */
542 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
544 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE)
545 get_pci_parity_status(dev, 1);
551 * Fucntion to retrieve the current parity status
555 static void edac_pci_dev_parity_test(struct pci_dev *dev)
561 /* stop any interrupts until we can acquire the status */
562 local_irq_save(flags);
564 /* read the STATUS register on this device */
565 status = get_pci_parity_status(dev, 0);
567 /* read the device TYPE, looking for bridges */
568 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
570 local_irq_restore(flags);
572 debugf4("PCI STATUS= 0x%04x %s\n", status, dev->dev.bus_id);
574 /* check the status reg for errors on boards NOT marked as broken
575 * if broken, we cannot trust any of the status bits
577 if (status && !dev->broken_parity_status) {
578 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {
579 edac_printk(KERN_CRIT, EDAC_PCI,
580 "Signaled System Error on %s\n",
582 atomic_inc(&pci_nonparity_count);
585 if (status & (PCI_STATUS_PARITY)) {
586 edac_printk(KERN_CRIT, EDAC_PCI,
587 "Master Data Parity Error on %s\n",
590 atomic_inc(&pci_parity_count);
593 if (status & (PCI_STATUS_DETECTED_PARITY)) {
594 edac_printk(KERN_CRIT, EDAC_PCI,
595 "Detected Parity Error on %s\n",
598 atomic_inc(&pci_parity_count);
603 debugf4("PCI HEADER TYPE= 0x%02x %s\n", header_type, dev->dev.bus_id);
605 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) {
606 /* On bridges, need to examine secondary status register */
607 status = get_pci_parity_status(dev, 1);
609 debugf4("PCI SEC_STATUS= 0x%04x %s\n", status, dev->dev.bus_id);
611 /* check the secondary status reg for errors,
612 * on NOT broken boards
614 if (status && !dev->broken_parity_status) {
615 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {
616 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
617 "Signaled System Error on %s\n",
619 atomic_inc(&pci_nonparity_count);
622 if (status & (PCI_STATUS_PARITY)) {
623 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
624 "Master Data Parity Error on "
625 "%s\n", pci_name(dev));
627 atomic_inc(&pci_parity_count);
630 if (status & (PCI_STATUS_DETECTED_PARITY)) {
631 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
632 "Detected Parity Error on %s\n",
635 atomic_inc(&pci_parity_count);
641 /* reduce some complexity in definition of the iterator */
642 typedef void (*pci_parity_check_fn_t) (struct pci_dev *dev);
645 * pci_dev parity list iterator
646 * Scan the PCI device list for one pass, looking for SERRORs
647 * Master Parity ERRORS or Parity ERRORs on primary or secondary devices
649 static inline void edac_pci_dev_parity_iterator(pci_parity_check_fn_t fn)
651 struct pci_dev *dev = NULL;
653 /* request for kernel access to the next PCI device, if any,
654 * and while we are looking at it have its reference count
655 * bumped until we are done with it
657 while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
663 * edac_pci_do_parity_check
665 * performs the actual PCI parity check operation
667 void edac_pci_do_parity_check(void)
671 debugf3("%s()\n", __func__);
673 /* if policy has PCI check off, leave now */
674 if (!check_pci_errors)
677 before_count = atomic_read(&pci_parity_count);
679 /* scan all PCI devices looking for a Parity Error on devices and
681 * The iterator calls pci_get_device() which might sleep, thus
682 * we cannot disable interrupts in this scan.
684 edac_pci_dev_parity_iterator(edac_pci_dev_parity_test);
686 /* Only if operator has selected panic on PCI Error */
687 if (edac_pci_get_panic_on_pe()) {
688 /* If the count is different 'after' from 'before' */
689 if (before_count != atomic_read(&pci_parity_count))
690 panic("EDAC: PCI Parity Error");
695 * edac_pci_clear_parity_errors
697 * function to perform an iteration over the PCI devices
698 * and clearn their current status
700 void edac_pci_clear_parity_errors(void)
702 /* Clear any PCI bus parity errors that devices initially have logged
703 * in their registers.
705 edac_pci_dev_parity_iterator(edac_pci_dev_parity_clear);
711 * Called to handle a PARITY ERROR event
713 void edac_pci_handle_pe(struct edac_pci_ctl_info *pci, const char *msg)
716 /* global PE counter incremented by edac_pci_do_parity_check() */
717 atomic_inc(&pci->counters.pe_count);
719 if (edac_pci_get_log_pe())
720 edac_pci_printk(pci, KERN_WARNING,
721 "Parity Error ctl: %s %d: %s\n",
722 pci->ctl_name, pci->pci_idx, msg);
725 * poke all PCI devices and see which one is the troublemaker
726 * panic() is called if set
728 edac_pci_do_parity_check();
730 EXPORT_SYMBOL_GPL(edac_pci_handle_pe);
734 * edac_pci_handle_npe
736 * Called to handle a NON-PARITY ERROR event
738 void edac_pci_handle_npe(struct edac_pci_ctl_info *pci, const char *msg)
741 /* global NPE counter incremented by edac_pci_do_parity_check() */
742 atomic_inc(&pci->counters.npe_count);
744 if (edac_pci_get_log_npe())
745 edac_pci_printk(pci, KERN_WARNING,
746 "Non-Parity Error ctl: %s %d: %s\n",
747 pci->ctl_name, pci->pci_idx, msg);
750 * poke all PCI devices and see which one is the troublemaker
751 * panic() is called if set
753 edac_pci_do_parity_check();
755 EXPORT_SYMBOL_GPL(edac_pci_handle_npe);
758 * Define the PCI parameter to the module
760 module_param(check_pci_errors, int, 0644);
761 MODULE_PARM_DESC(check_pci_errors,
762 "Check for PCI bus parity errors: 0=off 1=on");
763 module_param(edac_pci_panic_on_pe, int, 0644);
764 MODULE_PARM_DESC(edac_pci_panic_on_pe,
765 "Panic on PCI Bus Parity error: 0=off 1=on");
767 #endif /* CONFIG_PCI */