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>
15 #include "edac_module.h"
17 #define EDAC_PCI_SYMLINK "device"
19 /* data variables exported via sysfs */
20 static int check_pci_errors; /* default NO check PCI parity */
21 static int edac_pci_panic_on_pe; /* default NO panic on PCI Parity */
22 static int edac_pci_log_pe = 1; /* log PCI parity errors */
23 static int edac_pci_log_npe = 1; /* log PCI non-parity error errors */
24 static int edac_pci_poll_msec = 1000; /* one second workq period */
26 static atomic_t pci_parity_count = ATOMIC_INIT(0);
27 static atomic_t pci_nonparity_count = ATOMIC_INIT(0);
29 static struct kobject *edac_pci_top_main_kobj;
30 static atomic_t edac_pci_sysfs_refcount = ATOMIC_INIT(0);
32 /* getter functions for the data variables */
33 int edac_pci_get_check_errors(void)
35 return check_pci_errors;
38 static int edac_pci_get_log_pe(void)
40 return edac_pci_log_pe;
43 static int edac_pci_get_log_npe(void)
45 return edac_pci_log_npe;
48 static int edac_pci_get_panic_on_pe(void)
50 return edac_pci_panic_on_pe;
53 int edac_pci_get_poll_msec(void)
55 return edac_pci_poll_msec;
58 /**************************** EDAC PCI sysfs instance *******************/
59 static ssize_t instance_pe_count_show(struct edac_pci_ctl_info *pci, char *data)
61 return sprintf(data, "%u\n", atomic_read(&pci->counters.pe_count));
64 static ssize_t instance_npe_count_show(struct edac_pci_ctl_info *pci,
67 return sprintf(data, "%u\n", atomic_read(&pci->counters.npe_count));
70 #define to_instance(k) container_of(k, struct edac_pci_ctl_info, kobj)
71 #define to_instance_attr(a) container_of(a, struct instance_attribute, attr)
73 /* DEVICE instance kobject release() function */
74 static void edac_pci_instance_release(struct kobject *kobj)
76 struct edac_pci_ctl_info *pci;
80 /* Form pointer to containing struct, the pci control struct */
81 pci = to_instance(kobj);
83 /* decrement reference count on top main kobj */
84 kobject_put(edac_pci_top_main_kobj);
86 kfree(pci); /* Free the control struct */
89 /* instance specific attribute structure */
90 struct instance_attribute {
91 struct attribute attr;
92 ssize_t(*show) (struct edac_pci_ctl_info *, char *);
93 ssize_t(*store) (struct edac_pci_ctl_info *, const char *, size_t);
96 /* Function to 'show' fields from the edac_pci 'instance' structure */
97 static ssize_t edac_pci_instance_show(struct kobject *kobj,
98 struct attribute *attr, char *buffer)
100 struct edac_pci_ctl_info *pci = to_instance(kobj);
101 struct instance_attribute *instance_attr = to_instance_attr(attr);
103 if (instance_attr->show)
104 return instance_attr->show(pci, buffer);
108 /* Function to 'store' fields into the edac_pci 'instance' structure */
109 static ssize_t edac_pci_instance_store(struct kobject *kobj,
110 struct attribute *attr,
111 const char *buffer, size_t count)
113 struct edac_pci_ctl_info *pci = to_instance(kobj);
114 struct instance_attribute *instance_attr = to_instance_attr(attr);
116 if (instance_attr->store)
117 return instance_attr->store(pci, buffer, count);
122 static const struct sysfs_ops pci_instance_ops = {
123 .show = edac_pci_instance_show,
124 .store = edac_pci_instance_store
127 #define INSTANCE_ATTR(_name, _mode, _show, _store) \
128 static struct instance_attribute attr_instance_##_name = { \
129 .attr = {.name = __stringify(_name), .mode = _mode }, \
134 INSTANCE_ATTR(pe_count, S_IRUGO, instance_pe_count_show, NULL);
135 INSTANCE_ATTR(npe_count, S_IRUGO, instance_npe_count_show, NULL);
137 /* pci instance attributes */
138 static struct attribute *pci_instance_attrs[] = {
139 &attr_instance_pe_count.attr,
140 &attr_instance_npe_count.attr,
143 ATTRIBUTE_GROUPS(pci_instance);
145 /* the ktype for a pci instance */
146 static struct kobj_type ktype_pci_instance = {
147 .release = edac_pci_instance_release,
148 .sysfs_ops = &pci_instance_ops,
149 .default_groups = pci_instance_groups,
153 * edac_pci_create_instance_kobj
155 * construct one EDAC PCI instance's kobject for use
157 static int edac_pci_create_instance_kobj(struct edac_pci_ctl_info *pci, int idx)
159 struct kobject *main_kobj;
164 /* First bump the ref count on the top main kobj, which will
165 * track the number of PCI instances we have, and thus nest
166 * properly on keeping the module loaded
168 main_kobj = kobject_get(edac_pci_top_main_kobj);
174 /* And now register this new kobject under the main kobj */
175 err = kobject_init_and_add(&pci->kobj, &ktype_pci_instance,
176 edac_pci_top_main_kobj, "pci%d", idx);
178 edac_dbg(2, "failed to register instance pci%d\n", idx);
179 kobject_put(edac_pci_top_main_kobj);
183 kobject_uevent(&pci->kobj, KOBJ_ADD);
184 edac_dbg(1, "Register instance 'pci%d' kobject\n", idx);
188 /* Error unwind statck */
194 * edac_pci_unregister_sysfs_instance_kobj
196 * unregister the kobj for the EDAC PCI instance
198 static void edac_pci_unregister_sysfs_instance_kobj(
199 struct edac_pci_ctl_info *pci)
203 /* Unregister the instance kobject and allow its release
204 * function release the main reference count and then
207 kobject_put(&pci->kobj);
210 /***************************** EDAC PCI sysfs root **********************/
211 #define to_edacpci(k) container_of(k, struct edac_pci_ctl_info, kobj)
212 #define to_edacpci_attr(a) container_of(a, struct edac_pci_attr, attr)
214 /* simple show/store functions for attributes */
215 static ssize_t edac_pci_int_show(void *ptr, char *buffer)
218 return sprintf(buffer, "%d\n", *value);
221 static ssize_t edac_pci_int_store(void *ptr, const char *buffer, size_t count)
225 if (isdigit(*buffer))
226 *value = simple_strtoul(buffer, NULL, 0);
231 struct edac_pci_dev_attribute {
232 struct attribute attr;
234 ssize_t(*show) (void *, char *);
235 ssize_t(*store) (void *, const char *, size_t);
238 /* Set of show/store abstract level functions for PCI Parity object */
239 static ssize_t edac_pci_dev_show(struct kobject *kobj, struct attribute *attr,
242 struct edac_pci_dev_attribute *edac_pci_dev;
243 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;
245 if (edac_pci_dev->show)
246 return edac_pci_dev->show(edac_pci_dev->value, buffer);
250 static ssize_t edac_pci_dev_store(struct kobject *kobj,
251 struct attribute *attr, const char *buffer,
254 struct edac_pci_dev_attribute *edac_pci_dev;
255 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;
257 if (edac_pci_dev->store)
258 return edac_pci_dev->store(edac_pci_dev->value, buffer, count);
262 static const struct sysfs_ops edac_pci_sysfs_ops = {
263 .show = edac_pci_dev_show,
264 .store = edac_pci_dev_store
267 #define EDAC_PCI_ATTR(_name,_mode,_show,_store) \
268 static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
269 .attr = {.name = __stringify(_name), .mode = _mode }, \
275 #define EDAC_PCI_STRING_ATTR(_name,_data,_mode,_show,_store) \
276 static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
277 .attr = {.name = __stringify(_name), .mode = _mode }, \
283 /* PCI Parity control files */
284 EDAC_PCI_ATTR(check_pci_errors, S_IRUGO | S_IWUSR, edac_pci_int_show,
286 EDAC_PCI_ATTR(edac_pci_log_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,
288 EDAC_PCI_ATTR(edac_pci_log_npe, S_IRUGO | S_IWUSR, edac_pci_int_show,
290 EDAC_PCI_ATTR(edac_pci_panic_on_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,
292 EDAC_PCI_ATTR(pci_parity_count, S_IRUGO, edac_pci_int_show, NULL);
293 EDAC_PCI_ATTR(pci_nonparity_count, S_IRUGO, edac_pci_int_show, NULL);
295 /* Base Attributes of the memory ECC object */
296 static struct attribute *edac_pci_attrs[] = {
297 &edac_pci_attr_check_pci_errors.attr,
298 &edac_pci_attr_edac_pci_log_pe.attr,
299 &edac_pci_attr_edac_pci_log_npe.attr,
300 &edac_pci_attr_edac_pci_panic_on_pe.attr,
301 &edac_pci_attr_pci_parity_count.attr,
302 &edac_pci_attr_pci_nonparity_count.attr,
305 ATTRIBUTE_GROUPS(edac_pci);
308 * edac_pci_release_main_kobj
310 * This release function is called when the reference count to the
311 * passed kobj goes to zero.
313 * This kobj is the 'main' kobject that EDAC PCI instances
314 * link to, and thus provide for proper nesting counts
316 static void edac_pci_release_main_kobj(struct kobject *kobj)
318 edac_dbg(0, "here to module_put(THIS_MODULE)\n");
322 /* last reference to top EDAC PCI kobject has been removed,
323 * NOW release our ref count on the core module
325 module_put(THIS_MODULE);
328 /* ktype struct for the EDAC PCI main kobj */
329 static struct kobj_type ktype_edac_pci_main_kobj = {
330 .release = edac_pci_release_main_kobj,
331 .sysfs_ops = &edac_pci_sysfs_ops,
332 .default_groups = edac_pci_groups,
336 * edac_pci_main_kobj_setup: Setup the sysfs for EDAC PCI attributes.
338 static int edac_pci_main_kobj_setup(void)
341 struct bus_type *edac_subsys;
345 /* check and count if we have already created the main kobject */
346 if (atomic_inc_return(&edac_pci_sysfs_refcount) != 1)
349 /* First time, so create the main kobject and its
350 * controls and attributes
352 edac_subsys = edac_get_sysfs_subsys();
354 /* Bump the reference count on this module to ensure the
355 * modules isn't unloaded until we deconstruct the top
356 * level main kobj for EDAC PCI
358 if (!try_module_get(THIS_MODULE)) {
359 edac_dbg(1, "try_module_get() failed\n");
361 goto decrement_count_fail;
364 edac_pci_top_main_kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
365 if (!edac_pci_top_main_kobj) {
366 edac_dbg(1, "Failed to allocate\n");
371 /* Instanstiate the pci object */
372 err = kobject_init_and_add(edac_pci_top_main_kobj,
373 &ktype_edac_pci_main_kobj,
374 &edac_subsys->dev_root->kobj, "pci");
376 edac_dbg(1, "Failed to register '.../edac/pci'\n");
377 goto kobject_init_and_add_fail;
380 /* At this point, to 'release' the top level kobject
381 * for EDAC PCI, then edac_pci_main_kobj_teardown()
382 * must be used, for resources to be cleaned up properly
384 kobject_uevent(edac_pci_top_main_kobj, KOBJ_ADD);
385 edac_dbg(1, "Registered '.../edac/pci' kobject\n");
389 /* Error unwind statck */
390 kobject_init_and_add_fail:
391 kobject_put(edac_pci_top_main_kobj);
394 module_put(THIS_MODULE);
396 decrement_count_fail:
397 /* if are on this error exit, nothing to tear down */
398 atomic_dec(&edac_pci_sysfs_refcount);
404 * edac_pci_main_kobj_teardown()
406 * if no longer linked (needed) remove the top level EDAC PCI
407 * kobject with its controls and attributes
409 static void edac_pci_main_kobj_teardown(void)
413 /* Decrement the count and only if no more controller instances
414 * are connected perform the unregisteration of the top level
417 if (atomic_dec_return(&edac_pci_sysfs_refcount) == 0) {
418 edac_dbg(0, "called kobject_put on main kobj\n");
419 kobject_put(edac_pci_top_main_kobj);
423 int edac_pci_create_sysfs(struct edac_pci_ctl_info *pci)
426 struct kobject *edac_kobj = &pci->kobj;
428 edac_dbg(0, "idx=%d\n", pci->pci_idx);
430 /* create the top main EDAC PCI kobject, IF needed */
431 err = edac_pci_main_kobj_setup();
435 /* Create this instance's kobject under the MAIN kobject */
436 err = edac_pci_create_instance_kobj(pci, pci->pci_idx);
438 goto unregister_cleanup;
440 err = sysfs_create_link(edac_kobj, &pci->dev->kobj, EDAC_PCI_SYMLINK);
442 edac_dbg(0, "sysfs_create_link() returned err= %d\n", err);
448 /* Error unwind stack */
450 edac_pci_unregister_sysfs_instance_kobj(pci);
453 edac_pci_main_kobj_teardown();
458 void edac_pci_remove_sysfs(struct edac_pci_ctl_info *pci)
460 edac_dbg(0, "index=%d\n", pci->pci_idx);
462 /* Remove the symlink */
463 sysfs_remove_link(&pci->kobj, EDAC_PCI_SYMLINK);
465 /* remove this PCI instance's sysfs entries */
466 edac_pci_unregister_sysfs_instance_kobj(pci);
468 /* Call the main unregister function, which will determine
469 * if this 'pci' is the last instance.
470 * If it is, the main kobject will be unregistered as a result
472 edac_dbg(0, "calling edac_pci_main_kobj_teardown()\n");
473 edac_pci_main_kobj_teardown();
476 /************************ PCI error handling *************************/
477 static u16 get_pci_parity_status(struct pci_dev *dev, int secondary)
482 where = secondary ? PCI_SEC_STATUS : PCI_STATUS;
483 pci_read_config_word(dev, where, &status);
485 /* If we get back 0xFFFF then we must suspect that the card has been
486 * pulled but the Linux PCI layer has not yet finished cleaning up.
487 * We don't want to report on such devices
490 if (status == 0xFFFF) {
493 pci_read_config_dword(dev, 0, &sanity);
495 if (sanity == 0xFFFFFFFF)
499 status &= PCI_STATUS_DETECTED_PARITY | PCI_STATUS_SIG_SYSTEM_ERROR |
503 /* reset only the bits we are interested in */
504 pci_write_config_word(dev, where, status);
510 /* Clear any PCI parity errors logged by this device. */
511 static void edac_pci_dev_parity_clear(struct pci_dev *dev)
515 get_pci_parity_status(dev, 0);
517 /* read the device TYPE, looking for bridges */
518 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
520 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE)
521 get_pci_parity_status(dev, 1);
527 * Function to retrieve the current parity status
531 static void edac_pci_dev_parity_test(struct pci_dev *dev)
537 /* stop any interrupts until we can acquire the status */
538 local_irq_save(flags);
540 /* read the STATUS register on this device */
541 status = 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 local_irq_restore(flags);
548 edac_dbg(4, "PCI STATUS= 0x%04x %s\n", status, dev_name(&dev->dev));
550 /* check the status reg for errors on boards NOT marked as broken
551 * if broken, we cannot trust any of the status bits
553 if (status && !dev->broken_parity_status) {
554 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {
555 edac_printk(KERN_CRIT, EDAC_PCI,
556 "Signaled System Error on %s\n",
558 atomic_inc(&pci_nonparity_count);
561 if (status & (PCI_STATUS_PARITY)) {
562 edac_printk(KERN_CRIT, EDAC_PCI,
563 "Master Data Parity Error on %s\n",
566 atomic_inc(&pci_parity_count);
569 if (status & (PCI_STATUS_DETECTED_PARITY)) {
570 edac_printk(KERN_CRIT, EDAC_PCI,
571 "Detected Parity Error on %s\n",
574 atomic_inc(&pci_parity_count);
579 edac_dbg(4, "PCI HEADER TYPE= 0x%02x %s\n",
580 header_type, dev_name(&dev->dev));
582 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) {
583 /* On bridges, need to examine secondary status register */
584 status = get_pci_parity_status(dev, 1);
586 edac_dbg(4, "PCI SEC_STATUS= 0x%04x %s\n",
587 status, dev_name(&dev->dev));
589 /* check the secondary status reg for errors,
590 * on NOT broken boards
592 if (status && !dev->broken_parity_status) {
593 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {
594 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
595 "Signaled System Error on %s\n",
597 atomic_inc(&pci_nonparity_count);
600 if (status & (PCI_STATUS_PARITY)) {
601 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
602 "Master Data Parity Error on "
603 "%s\n", pci_name(dev));
605 atomic_inc(&pci_parity_count);
608 if (status & (PCI_STATUS_DETECTED_PARITY)) {
609 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
610 "Detected Parity Error on %s\n",
613 atomic_inc(&pci_parity_count);
619 /* reduce some complexity in definition of the iterator */
620 typedef void (*pci_parity_check_fn_t) (struct pci_dev *dev);
623 * pci_dev parity list iterator
625 * Scan the PCI device list looking for SERRORs, Master Parity ERRORS or
626 * Parity ERRORs on primary or secondary devices.
628 static inline void edac_pci_dev_parity_iterator(pci_parity_check_fn_t fn)
630 struct pci_dev *dev = NULL;
632 for_each_pci_dev(dev)
637 * edac_pci_do_parity_check
639 * performs the actual PCI parity check operation
641 void edac_pci_do_parity_check(void)
647 /* if policy has PCI check off, leave now */
648 if (!check_pci_errors)
651 before_count = atomic_read(&pci_parity_count);
653 /* scan all PCI devices looking for a Parity Error on devices and
655 * The iterator calls pci_get_device() which might sleep, thus
656 * we cannot disable interrupts in this scan.
658 edac_pci_dev_parity_iterator(edac_pci_dev_parity_test);
660 /* Only if operator has selected panic on PCI Error */
661 if (edac_pci_get_panic_on_pe()) {
662 /* If the count is different 'after' from 'before' */
663 if (before_count != atomic_read(&pci_parity_count))
664 panic("EDAC: PCI Parity Error");
669 * edac_pci_clear_parity_errors
671 * function to perform an iteration over the PCI devices
672 * and clearn their current status
674 void edac_pci_clear_parity_errors(void)
676 /* Clear any PCI bus parity errors that devices initially have logged
677 * in their registers.
679 edac_pci_dev_parity_iterator(edac_pci_dev_parity_clear);
685 * Called to handle a PARITY ERROR event
687 void edac_pci_handle_pe(struct edac_pci_ctl_info *pci, const char *msg)
690 /* global PE counter incremented by edac_pci_do_parity_check() */
691 atomic_inc(&pci->counters.pe_count);
693 if (edac_pci_get_log_pe())
694 edac_pci_printk(pci, KERN_WARNING,
695 "Parity Error ctl: %s %d: %s\n",
696 pci->ctl_name, pci->pci_idx, msg);
699 * poke all PCI devices and see which one is the troublemaker
700 * panic() is called if set
702 edac_pci_do_parity_check();
704 EXPORT_SYMBOL_GPL(edac_pci_handle_pe);
708 * edac_pci_handle_npe
710 * Called to handle a NON-PARITY ERROR event
712 void edac_pci_handle_npe(struct edac_pci_ctl_info *pci, const char *msg)
715 /* global NPE counter incremented by edac_pci_do_parity_check() */
716 atomic_inc(&pci->counters.npe_count);
718 if (edac_pci_get_log_npe())
719 edac_pci_printk(pci, KERN_WARNING,
720 "Non-Parity Error ctl: %s %d: %s\n",
721 pci->ctl_name, pci->pci_idx, msg);
724 * poke all PCI devices and see which one is the troublemaker
725 * panic() is called if set
727 edac_pci_do_parity_check();
729 EXPORT_SYMBOL_GPL(edac_pci_handle_npe);
732 * Define the PCI parameter to the module
734 module_param(check_pci_errors, int, 0644);
735 MODULE_PARM_DESC(check_pci_errors,
736 "Check for PCI bus parity errors: 0=off 1=on");
737 module_param(edac_pci_panic_on_pe, int, 0644);
738 MODULE_PARM_DESC(edac_pci_panic_on_pe,
739 "Panic on PCI Bus Parity error: 0=off 1=on");