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 #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 instance_attribute *pci_instance_attr[] = {
139 &attr_instance_pe_count,
140 &attr_instance_npe_count,
144 /* the ktype for a pci instance */
145 static struct kobj_type ktype_pci_instance = {
146 .release = edac_pci_instance_release,
147 .sysfs_ops = &pci_instance_ops,
148 .default_attrs = (struct attribute **)pci_instance_attr,
152 * edac_pci_create_instance_kobj
154 * construct one EDAC PCI instance's kobject for use
156 static int edac_pci_create_instance_kobj(struct edac_pci_ctl_info *pci, int idx)
158 struct kobject *main_kobj;
163 /* First bump the ref count on the top main kobj, which will
164 * track the number of PCI instances we have, and thus nest
165 * properly on keeping the module loaded
167 main_kobj = kobject_get(edac_pci_top_main_kobj);
173 /* And now register this new kobject under the main kobj */
174 err = kobject_init_and_add(&pci->kobj, &ktype_pci_instance,
175 edac_pci_top_main_kobj, "pci%d", idx);
177 edac_dbg(2, "failed to register instance pci%d\n", idx);
178 kobject_put(edac_pci_top_main_kobj);
182 kobject_uevent(&pci->kobj, KOBJ_ADD);
183 edac_dbg(1, "Register instance 'pci%d' kobject\n", idx);
187 /* Error unwind statck */
193 * edac_pci_unregister_sysfs_instance_kobj
195 * unregister the kobj for the EDAC PCI instance
197 static void edac_pci_unregister_sysfs_instance_kobj(
198 struct edac_pci_ctl_info *pci)
202 /* Unregister the instance kobject and allow its release
203 * function release the main reference count and then
206 kobject_put(&pci->kobj);
209 /***************************** EDAC PCI sysfs root **********************/
210 #define to_edacpci(k) container_of(k, struct edac_pci_ctl_info, kobj)
211 #define to_edacpci_attr(a) container_of(a, struct edac_pci_attr, attr)
213 /* simple show/store functions for attributes */
214 static ssize_t edac_pci_int_show(void *ptr, char *buffer)
217 return sprintf(buffer, "%d\n", *value);
220 static ssize_t edac_pci_int_store(void *ptr, const char *buffer, size_t count)
224 if (isdigit(*buffer))
225 *value = simple_strtoul(buffer, NULL, 0);
230 struct edac_pci_dev_attribute {
231 struct attribute attr;
233 ssize_t(*show) (void *, char *);
234 ssize_t(*store) (void *, const char *, size_t);
237 /* Set of show/store abstract level functions for PCI Parity object */
238 static ssize_t edac_pci_dev_show(struct kobject *kobj, struct attribute *attr,
241 struct edac_pci_dev_attribute *edac_pci_dev;
242 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;
244 if (edac_pci_dev->show)
245 return edac_pci_dev->show(edac_pci_dev->value, buffer);
249 static ssize_t edac_pci_dev_store(struct kobject *kobj,
250 struct attribute *attr, const char *buffer,
253 struct edac_pci_dev_attribute *edac_pci_dev;
254 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;
256 if (edac_pci_dev->store)
257 return edac_pci_dev->store(edac_pci_dev->value, buffer, count);
261 static const struct sysfs_ops edac_pci_sysfs_ops = {
262 .show = edac_pci_dev_show,
263 .store = edac_pci_dev_store
266 #define EDAC_PCI_ATTR(_name,_mode,_show,_store) \
267 static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
268 .attr = {.name = __stringify(_name), .mode = _mode }, \
274 #define EDAC_PCI_STRING_ATTR(_name,_data,_mode,_show,_store) \
275 static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \
276 .attr = {.name = __stringify(_name), .mode = _mode }, \
282 /* PCI Parity control files */
283 EDAC_PCI_ATTR(check_pci_errors, S_IRUGO | S_IWUSR, edac_pci_int_show,
285 EDAC_PCI_ATTR(edac_pci_log_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,
287 EDAC_PCI_ATTR(edac_pci_log_npe, S_IRUGO | S_IWUSR, edac_pci_int_show,
289 EDAC_PCI_ATTR(edac_pci_panic_on_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,
291 EDAC_PCI_ATTR(pci_parity_count, S_IRUGO, edac_pci_int_show, NULL);
292 EDAC_PCI_ATTR(pci_nonparity_count, S_IRUGO, edac_pci_int_show, NULL);
294 /* Base Attributes of the memory ECC object */
295 static struct edac_pci_dev_attribute *edac_pci_attr[] = {
296 &edac_pci_attr_check_pci_errors,
297 &edac_pci_attr_edac_pci_log_pe,
298 &edac_pci_attr_edac_pci_log_npe,
299 &edac_pci_attr_edac_pci_panic_on_pe,
300 &edac_pci_attr_pci_parity_count,
301 &edac_pci_attr_pci_nonparity_count,
306 * edac_pci_release_main_kobj
308 * This release function is called when the reference count to the
309 * passed kobj goes to zero.
311 * This kobj is the 'main' kobject that EDAC PCI instances
312 * link to, and thus provide for proper nesting counts
314 static void edac_pci_release_main_kobj(struct kobject *kobj)
316 edac_dbg(0, "here to module_put(THIS_MODULE)\n");
320 /* last reference to top EDAC PCI kobject has been removed,
321 * NOW release our ref count on the core module
323 module_put(THIS_MODULE);
326 /* ktype struct for the EDAC PCI main kobj */
327 static struct kobj_type ktype_edac_pci_main_kobj = {
328 .release = edac_pci_release_main_kobj,
329 .sysfs_ops = &edac_pci_sysfs_ops,
330 .default_attrs = (struct attribute **)edac_pci_attr,
334 * edac_pci_main_kobj_setup()
336 * setup the sysfs for EDAC PCI attributes
337 * assumes edac_subsys has already been initialized
339 static int edac_pci_main_kobj_setup(void)
342 struct bus_type *edac_subsys;
346 /* check and count if we have already created the main kobject */
347 if (atomic_inc_return(&edac_pci_sysfs_refcount) != 1)
350 /* First time, so create the main kobject and its
351 * controls and attributes
353 edac_subsys = edac_get_sysfs_subsys();
354 if (edac_subsys == NULL) {
355 edac_dbg(1, "no edac_subsys\n");
357 goto decrement_count_fail;
360 /* Bump the reference count on this module to ensure the
361 * modules isn't unloaded until we deconstruct the top
362 * level main kobj for EDAC PCI
364 if (!try_module_get(THIS_MODULE)) {
365 edac_dbg(1, "try_module_get() failed\n");
370 edac_pci_top_main_kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
371 if (!edac_pci_top_main_kobj) {
372 edac_dbg(1, "Failed to allocate\n");
377 /* Instanstiate the pci object */
378 err = kobject_init_and_add(edac_pci_top_main_kobj,
379 &ktype_edac_pci_main_kobj,
380 &edac_subsys->dev_root->kobj, "pci");
382 edac_dbg(1, "Failed to register '.../edac/pci'\n");
383 goto kobject_init_and_add_fail;
386 /* At this point, to 'release' the top level kobject
387 * for EDAC PCI, then edac_pci_main_kobj_teardown()
388 * must be used, for resources to be cleaned up properly
390 kobject_uevent(edac_pci_top_main_kobj, KOBJ_ADD);
391 edac_dbg(1, "Registered '.../edac/pci' kobject\n");
395 /* Error unwind statck */
396 kobject_init_and_add_fail:
397 kobject_put(edac_pci_top_main_kobj);
400 module_put(THIS_MODULE);
403 edac_put_sysfs_subsys();
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)
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 edac_dbg(0, "called kobject_put on main kobj\n");
428 kobject_put(edac_pci_top_main_kobj);
429 edac_put_sysfs_subsys();
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 edac_dbg(0, "idx=%d\n", 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 edac_dbg(0, "sysfs_create_link() returned err= %d\n", err);
464 /* Error unwind stack */
466 edac_pci_unregister_sysfs_instance_kobj(pci);
469 edac_pci_main_kobj_teardown();
475 * edac_pci_remove_sysfs
477 * remove the controls and attributes for this EDAC PCI device
479 void edac_pci_remove_sysfs(struct edac_pci_ctl_info *pci)
481 edac_dbg(0, "index=%d\n", pci->pci_idx);
483 /* Remove the symlink */
484 sysfs_remove_link(&pci->kobj, EDAC_PCI_SYMLINK);
486 /* remove this PCI instance's sysfs entries */
487 edac_pci_unregister_sysfs_instance_kobj(pci);
489 /* Call the main unregister function, which will determine
490 * if this 'pci' is the last instance.
491 * If it is, the main kobject will be unregistered as a result
493 edac_dbg(0, "calling edac_pci_main_kobj_teardown()\n");
494 edac_pci_main_kobj_teardown();
497 /************************ PCI error handling *************************/
498 static u16 get_pci_parity_status(struct pci_dev *dev, int secondary)
503 where = secondary ? PCI_SEC_STATUS : PCI_STATUS;
504 pci_read_config_word(dev, where, &status);
506 /* If we get back 0xFFFF then we must suspect that the card has been
507 * pulled but the Linux PCI layer has not yet finished cleaning up.
508 * We don't want to report on such devices
511 if (status == 0xFFFF) {
514 pci_read_config_dword(dev, 0, &sanity);
516 if (sanity == 0xFFFFFFFF)
520 status &= PCI_STATUS_DETECTED_PARITY | PCI_STATUS_SIG_SYSTEM_ERROR |
524 /* reset only the bits we are interested in */
525 pci_write_config_word(dev, where, status);
531 /* Clear any PCI parity errors logged by this device. */
532 static void edac_pci_dev_parity_clear(struct pci_dev *dev)
536 get_pci_parity_status(dev, 0);
538 /* read the device TYPE, looking for bridges */
539 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
541 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE)
542 get_pci_parity_status(dev, 1);
548 * Function to retrieve the current parity status
552 static void edac_pci_dev_parity_test(struct pci_dev *dev)
558 /* stop any interrupts until we can acquire the status */
559 local_irq_save(flags);
561 /* read the STATUS register on this device */
562 status = get_pci_parity_status(dev, 0);
564 /* read the device TYPE, looking for bridges */
565 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
567 local_irq_restore(flags);
569 edac_dbg(4, "PCI STATUS= 0x%04x %s\n", status, dev_name(&dev->dev));
571 /* check the status reg for errors on boards NOT marked as broken
572 * if broken, we cannot trust any of the status bits
574 if (status && !dev->broken_parity_status) {
575 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {
576 edac_printk(KERN_CRIT, EDAC_PCI,
577 "Signaled System Error on %s\n",
579 atomic_inc(&pci_nonparity_count);
582 if (status & (PCI_STATUS_PARITY)) {
583 edac_printk(KERN_CRIT, EDAC_PCI,
584 "Master Data Parity Error on %s\n",
587 atomic_inc(&pci_parity_count);
590 if (status & (PCI_STATUS_DETECTED_PARITY)) {
591 edac_printk(KERN_CRIT, EDAC_PCI,
592 "Detected Parity Error on %s\n",
595 atomic_inc(&pci_parity_count);
600 edac_dbg(4, "PCI HEADER TYPE= 0x%02x %s\n",
601 header_type, dev_name(&dev->dev));
603 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) {
604 /* On bridges, need to examine secondary status register */
605 status = get_pci_parity_status(dev, 1);
607 edac_dbg(4, "PCI SEC_STATUS= 0x%04x %s\n",
608 status, dev_name(&dev->dev));
610 /* check the secondary status reg for errors,
611 * on NOT broken boards
613 if (status && !dev->broken_parity_status) {
614 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {
615 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
616 "Signaled System Error on %s\n",
618 atomic_inc(&pci_nonparity_count);
621 if (status & (PCI_STATUS_PARITY)) {
622 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
623 "Master Data Parity Error on "
624 "%s\n", pci_name(dev));
626 atomic_inc(&pci_parity_count);
629 if (status & (PCI_STATUS_DETECTED_PARITY)) {
630 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "
631 "Detected Parity Error on %s\n",
634 atomic_inc(&pci_parity_count);
640 /* reduce some complexity in definition of the iterator */
641 typedef void (*pci_parity_check_fn_t) (struct pci_dev *dev);
644 * pci_dev parity list iterator
646 * Scan the PCI device list looking for SERRORs, Master Parity ERRORS or
647 * 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 for_each_pci_dev(dev)
658 * edac_pci_do_parity_check
660 * performs the actual PCI parity check operation
662 void edac_pci_do_parity_check(void)
668 /* if policy has PCI check off, leave now */
669 if (!check_pci_errors)
672 before_count = atomic_read(&pci_parity_count);
674 /* scan all PCI devices looking for a Parity Error on devices and
676 * The iterator calls pci_get_device() which might sleep, thus
677 * we cannot disable interrupts in this scan.
679 edac_pci_dev_parity_iterator(edac_pci_dev_parity_test);
681 /* Only if operator has selected panic on PCI Error */
682 if (edac_pci_get_panic_on_pe()) {
683 /* If the count is different 'after' from 'before' */
684 if (before_count != atomic_read(&pci_parity_count))
685 panic("EDAC: PCI Parity Error");
690 * edac_pci_clear_parity_errors
692 * function to perform an iteration over the PCI devices
693 * and clearn their current status
695 void edac_pci_clear_parity_errors(void)
697 /* Clear any PCI bus parity errors that devices initially have logged
698 * in their registers.
700 edac_pci_dev_parity_iterator(edac_pci_dev_parity_clear);
706 * Called to handle a PARITY ERROR event
708 void edac_pci_handle_pe(struct edac_pci_ctl_info *pci, const char *msg)
711 /* global PE counter incremented by edac_pci_do_parity_check() */
712 atomic_inc(&pci->counters.pe_count);
714 if (edac_pci_get_log_pe())
715 edac_pci_printk(pci, KERN_WARNING,
716 "Parity Error ctl: %s %d: %s\n",
717 pci->ctl_name, pci->pci_idx, msg);
720 * poke all PCI devices and see which one is the troublemaker
721 * panic() is called if set
723 edac_pci_do_parity_check();
725 EXPORT_SYMBOL_GPL(edac_pci_handle_pe);
729 * edac_pci_handle_npe
731 * Called to handle a NON-PARITY ERROR event
733 void edac_pci_handle_npe(struct edac_pci_ctl_info *pci, const char *msg)
736 /* global NPE counter incremented by edac_pci_do_parity_check() */
737 atomic_inc(&pci->counters.npe_count);
739 if (edac_pci_get_log_npe())
740 edac_pci_printk(pci, KERN_WARNING,
741 "Non-Parity Error ctl: %s %d: %s\n",
742 pci->ctl_name, pci->pci_idx, msg);
745 * poke all PCI devices and see which one is the troublemaker
746 * panic() is called if set
748 edac_pci_do_parity_check();
750 EXPORT_SYMBOL_GPL(edac_pci_handle_npe);
753 * Define the PCI parameter to the module
755 module_param(check_pci_errors, int, 0644);
756 MODULE_PARM_DESC(check_pci_errors,
757 "Check for PCI bus parity errors: 0=off 1=on");
758 module_param(edac_pci_panic_on_pe, int, 0644);
759 MODULE_PARM_DESC(edac_pci_panic_on_pe,
760 "Panic on PCI Bus Parity error: 0=off 1=on");