744 lines · c
1/*2 * (C) 2005, 2006 Linux Networx (http://lnxi.com)3 * This file may be distributed under the terms of the4 * GNU General Public License.5 *6 * Written Doug Thompson <norsk5@xmission.com>7 *8 */9#include <linux/module.h>10#include <linux/edac.h>11#include <linux/slab.h>12#include <linux/ctype.h>13 14#include "edac_pci.h"15#include "edac_module.h"16 17#define EDAC_PCI_SYMLINK "device"18 19/* data variables exported via sysfs */20static int check_pci_errors; /* default NO check PCI parity */21static int edac_pci_panic_on_pe; /* default NO panic on PCI Parity */22static int edac_pci_log_pe = 1; /* log PCI parity errors */23static int edac_pci_log_npe = 1; /* log PCI non-parity error errors */24static int edac_pci_poll_msec = 1000; /* one second workq period */25 26static atomic_t pci_parity_count = ATOMIC_INIT(0);27static atomic_t pci_nonparity_count = ATOMIC_INIT(0);28 29static struct kobject *edac_pci_top_main_kobj;30static atomic_t edac_pci_sysfs_refcount = ATOMIC_INIT(0);31 32/* getter functions for the data variables */33int edac_pci_get_check_errors(void)34{35 return check_pci_errors;36}37 38static int edac_pci_get_log_pe(void)39{40 return edac_pci_log_pe;41}42 43static int edac_pci_get_log_npe(void)44{45 return edac_pci_log_npe;46}47 48static int edac_pci_get_panic_on_pe(void)49{50 return edac_pci_panic_on_pe;51}52 53int edac_pci_get_poll_msec(void)54{55 return edac_pci_poll_msec;56}57 58/**************************** EDAC PCI sysfs instance *******************/59static ssize_t instance_pe_count_show(struct edac_pci_ctl_info *pci, char *data)60{61 return sprintf(data, "%u\n", atomic_read(&pci->counters.pe_count));62}63 64static ssize_t instance_npe_count_show(struct edac_pci_ctl_info *pci,65 char *data)66{67 return sprintf(data, "%u\n", atomic_read(&pci->counters.npe_count));68}69 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)72 73/* DEVICE instance kobject release() function */74static void edac_pci_instance_release(struct kobject *kobj)75{76 struct edac_pci_ctl_info *pci;77 78 edac_dbg(0, "\n");79 80 /* Form pointer to containing struct, the pci control struct */81 pci = to_instance(kobj);82 83 /* decrement reference count on top main kobj */84 kobject_put(edac_pci_top_main_kobj);85 86 kfree(pci); /* Free the control struct */87}88 89/* instance specific attribute structure */90struct 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);94};95 96/* Function to 'show' fields from the edac_pci 'instance' structure */97static ssize_t edac_pci_instance_show(struct kobject *kobj,98 struct attribute *attr, char *buffer)99{100 struct edac_pci_ctl_info *pci = to_instance(kobj);101 struct instance_attribute *instance_attr = to_instance_attr(attr);102 103 if (instance_attr->show)104 return instance_attr->show(pci, buffer);105 return -EIO;106}107 108/* Function to 'store' fields into the edac_pci 'instance' structure */109static ssize_t edac_pci_instance_store(struct kobject *kobj,110 struct attribute *attr,111 const char *buffer, size_t count)112{113 struct edac_pci_ctl_info *pci = to_instance(kobj);114 struct instance_attribute *instance_attr = to_instance_attr(attr);115 116 if (instance_attr->store)117 return instance_attr->store(pci, buffer, count);118 return -EIO;119}120 121/* fs_ops table */122static const struct sysfs_ops pci_instance_ops = {123 .show = edac_pci_instance_show,124 .store = edac_pci_instance_store125};126 127#define INSTANCE_ATTR(_name, _mode, _show, _store) \128static struct instance_attribute attr_instance_##_name = { \129 .attr = {.name = __stringify(_name), .mode = _mode }, \130 .show = _show, \131 .store = _store, \132};133 134INSTANCE_ATTR(pe_count, S_IRUGO, instance_pe_count_show, NULL);135INSTANCE_ATTR(npe_count, S_IRUGO, instance_npe_count_show, NULL);136 137/* pci instance attributes */138static struct attribute *pci_instance_attrs[] = {139 &attr_instance_pe_count.attr,140 &attr_instance_npe_count.attr,141 NULL142};143ATTRIBUTE_GROUPS(pci_instance);144 145/* the ktype for a pci instance */146static struct kobj_type ktype_pci_instance = {147 .release = edac_pci_instance_release,148 .sysfs_ops = &pci_instance_ops,149 .default_groups = pci_instance_groups,150};151 152/*153 * edac_pci_create_instance_kobj154 *155 * construct one EDAC PCI instance's kobject for use156 */157static int edac_pci_create_instance_kobj(struct edac_pci_ctl_info *pci, int idx)158{159 struct kobject *main_kobj;160 int err;161 162 edac_dbg(0, "\n");163 164 /* First bump the ref count on the top main kobj, which will165 * track the number of PCI instances we have, and thus nest166 * properly on keeping the module loaded167 */168 main_kobj = kobject_get(edac_pci_top_main_kobj);169 if (!main_kobj) {170 err = -ENODEV;171 goto error_out;172 }173 174 /* 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);177 if (err != 0) {178 edac_dbg(2, "failed to register instance pci%d\n", idx);179 kobject_put(edac_pci_top_main_kobj);180 goto error_out;181 }182 183 kobject_uevent(&pci->kobj, KOBJ_ADD);184 edac_dbg(1, "Register instance 'pci%d' kobject\n", idx);185 186 return 0;187 188 /* Error unwind statck */189error_out:190 return err;191}192 193/*194 * edac_pci_unregister_sysfs_instance_kobj195 *196 * unregister the kobj for the EDAC PCI instance197 */198static void edac_pci_unregister_sysfs_instance_kobj(199 struct edac_pci_ctl_info *pci)200{201 edac_dbg(0, "\n");202 203 /* Unregister the instance kobject and allow its release204 * function release the main reference count and then205 * kfree the memory206 */207 kobject_put(&pci->kobj);208}209 210/***************************** 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)213 214/* simple show/store functions for attributes */215static ssize_t edac_pci_int_show(void *ptr, char *buffer)216{217 int *value = ptr;218 return sprintf(buffer, "%d\n", *value);219}220 221static ssize_t edac_pci_int_store(void *ptr, const char *buffer, size_t count)222{223 int *value = ptr;224 225 if (isdigit(*buffer))226 *value = simple_strtoul(buffer, NULL, 0);227 228 return count;229}230 231struct edac_pci_dev_attribute {232 struct attribute attr;233 void *value;234 ssize_t(*show) (void *, char *);235 ssize_t(*store) (void *, const char *, size_t);236};237 238/* Set of show/store abstract level functions for PCI Parity object */239static ssize_t edac_pci_dev_show(struct kobject *kobj, struct attribute *attr,240 char *buffer)241{242 struct edac_pci_dev_attribute *edac_pci_dev;243 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;244 245 if (edac_pci_dev->show)246 return edac_pci_dev->show(edac_pci_dev->value, buffer);247 return -EIO;248}249 250static ssize_t edac_pci_dev_store(struct kobject *kobj,251 struct attribute *attr, const char *buffer,252 size_t count)253{254 struct edac_pci_dev_attribute *edac_pci_dev;255 edac_pci_dev = (struct edac_pci_dev_attribute *)attr;256 257 if (edac_pci_dev->store)258 return edac_pci_dev->store(edac_pci_dev->value, buffer, count);259 return -EIO;260}261 262static const struct sysfs_ops edac_pci_sysfs_ops = {263 .show = edac_pci_dev_show,264 .store = edac_pci_dev_store265};266 267#define EDAC_PCI_ATTR(_name,_mode,_show,_store) \268static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \269 .attr = {.name = __stringify(_name), .mode = _mode }, \270 .value = &_name, \271 .show = _show, \272 .store = _store, \273};274 275#define EDAC_PCI_STRING_ATTR(_name,_data,_mode,_show,_store) \276static struct edac_pci_dev_attribute edac_pci_attr_##_name = { \277 .attr = {.name = __stringify(_name), .mode = _mode }, \278 .value = _data, \279 .show = _show, \280 .store = _store, \281};282 283/* PCI Parity control files */284EDAC_PCI_ATTR(check_pci_errors, S_IRUGO | S_IWUSR, edac_pci_int_show,285 edac_pci_int_store);286EDAC_PCI_ATTR(edac_pci_log_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,287 edac_pci_int_store);288EDAC_PCI_ATTR(edac_pci_log_npe, S_IRUGO | S_IWUSR, edac_pci_int_show,289 edac_pci_int_store);290EDAC_PCI_ATTR(edac_pci_panic_on_pe, S_IRUGO | S_IWUSR, edac_pci_int_show,291 edac_pci_int_store);292EDAC_PCI_ATTR(pci_parity_count, S_IRUGO, edac_pci_int_show, NULL);293EDAC_PCI_ATTR(pci_nonparity_count, S_IRUGO, edac_pci_int_show, NULL);294 295/* Base Attributes of the memory ECC object */296static 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,303 NULL,304};305ATTRIBUTE_GROUPS(edac_pci);306 307/*308 * edac_pci_release_main_kobj309 *310 * This release function is called when the reference count to the311 * passed kobj goes to zero.312 *313 * This kobj is the 'main' kobject that EDAC PCI instances314 * link to, and thus provide for proper nesting counts315 */316static void edac_pci_release_main_kobj(struct kobject *kobj)317{318 edac_dbg(0, "here to module_put(THIS_MODULE)\n");319 320 kfree(kobj);321 322 /* last reference to top EDAC PCI kobject has been removed,323 * NOW release our ref count on the core module324 */325 module_put(THIS_MODULE);326}327 328/* ktype struct for the EDAC PCI main kobj */329static 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,333};334 335/**336 * edac_pci_main_kobj_setup: Setup the sysfs for EDAC PCI attributes.337 */338static int edac_pci_main_kobj_setup(void)339{340 int err = -ENODEV;341 const struct bus_type *edac_subsys;342 struct device *dev_root;343 344 edac_dbg(0, "\n");345 346 /* check and count if we have already created the main kobject */347 if (atomic_inc_return(&edac_pci_sysfs_refcount) != 1)348 return 0;349 350 /* First time, so create the main kobject and its351 * controls and attributes352 */353 edac_subsys = edac_get_sysfs_subsys();354 355 /* Bump the reference count on this module to ensure the356 * modules isn't unloaded until we deconstruct the top357 * level main kobj for EDAC PCI358 */359 if (!try_module_get(THIS_MODULE)) {360 edac_dbg(1, "try_module_get() failed\n");361 goto decrement_count_fail;362 }363 364 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");367 err = -ENOMEM;368 goto kzalloc_fail;369 }370 371 /* Instanstiate the pci object */372 dev_root = bus_get_dev_root(edac_subsys);373 if (dev_root) {374 err = kobject_init_and_add(edac_pci_top_main_kobj,375 &ktype_edac_pci_main_kobj,376 &dev_root->kobj, "pci");377 put_device(dev_root);378 }379 if (err) {380 edac_dbg(1, "Failed to register '.../edac/pci'\n");381 goto kobject_init_and_add_fail;382 }383 384 /* At this point, to 'release' the top level kobject385 * for EDAC PCI, then edac_pci_main_kobj_teardown()386 * must be used, for resources to be cleaned up properly387 */388 kobject_uevent(edac_pci_top_main_kobj, KOBJ_ADD);389 edac_dbg(1, "Registered '.../edac/pci' kobject\n");390 391 return 0;392 393 /* Error unwind statck */394kobject_init_and_add_fail:395 kobject_put(edac_pci_top_main_kobj);396 397kzalloc_fail:398 module_put(THIS_MODULE);399 400decrement_count_fail:401 /* if are on this error exit, nothing to tear down */402 atomic_dec(&edac_pci_sysfs_refcount);403 404 return err;405}406 407/*408 * edac_pci_main_kobj_teardown()409 *410 * if no longer linked (needed) remove the top level EDAC PCI411 * kobject with its controls and attributes412 */413static void edac_pci_main_kobj_teardown(void)414{415 edac_dbg(0, "\n");416 417 /* Decrement the count and only if no more controller instances418 * are connected perform the unregisteration of the top level419 * main kobj420 */421 if (atomic_dec_return(&edac_pci_sysfs_refcount) == 0) {422 edac_dbg(0, "called kobject_put on main kobj\n");423 kobject_put(edac_pci_top_main_kobj);424 }425}426 427int edac_pci_create_sysfs(struct edac_pci_ctl_info *pci)428{429 int err;430 struct kobject *edac_kobj = &pci->kobj;431 432 edac_dbg(0, "idx=%d\n", pci->pci_idx);433 434 /* create the top main EDAC PCI kobject, IF needed */435 err = edac_pci_main_kobj_setup();436 if (err)437 return err;438 439 /* Create this instance's kobject under the MAIN kobject */440 err = edac_pci_create_instance_kobj(pci, pci->pci_idx);441 if (err)442 goto unregister_cleanup;443 444 err = sysfs_create_link(edac_kobj, &pci->dev->kobj, EDAC_PCI_SYMLINK);445 if (err) {446 edac_dbg(0, "sysfs_create_link() returned err= %d\n", err);447 goto symlink_fail;448 }449 450 return 0;451 452 /* Error unwind stack */453symlink_fail:454 edac_pci_unregister_sysfs_instance_kobj(pci);455 456unregister_cleanup:457 edac_pci_main_kobj_teardown();458 459 return err;460}461 462void edac_pci_remove_sysfs(struct edac_pci_ctl_info *pci)463{464 edac_dbg(0, "index=%d\n", pci->pci_idx);465 466 /* Remove the symlink */467 sysfs_remove_link(&pci->kobj, EDAC_PCI_SYMLINK);468 469 /* remove this PCI instance's sysfs entries */470 edac_pci_unregister_sysfs_instance_kobj(pci);471 472 /* Call the main unregister function, which will determine473 * if this 'pci' is the last instance.474 * If it is, the main kobject will be unregistered as a result475 */476 edac_dbg(0, "calling edac_pci_main_kobj_teardown()\n");477 edac_pci_main_kobj_teardown();478}479 480/************************ PCI error handling *************************/481static u16 get_pci_parity_status(struct pci_dev *dev, int secondary)482{483 int where;484 u16 status;485 486 where = secondary ? PCI_SEC_STATUS : PCI_STATUS;487 pci_read_config_word(dev, where, &status);488 489 /* If we get back 0xFFFF then we must suspect that the card has been490 * pulled but the Linux PCI layer has not yet finished cleaning up.491 * We don't want to report on such devices492 */493 494 if (status == 0xFFFF) {495 u32 sanity;496 497 pci_read_config_dword(dev, 0, &sanity);498 499 if (sanity == 0xFFFFFFFF)500 return 0;501 }502 503 status &= PCI_STATUS_DETECTED_PARITY | PCI_STATUS_SIG_SYSTEM_ERROR |504 PCI_STATUS_PARITY;505 506 if (status)507 /* reset only the bits we are interested in */508 pci_write_config_word(dev, where, status);509 510 return status;511}512 513 514/* Clear any PCI parity errors logged by this device. */515static void edac_pci_dev_parity_clear(struct pci_dev *dev)516{517 u8 header_type;518 519 get_pci_parity_status(dev, 0);520 521 /* read the device TYPE, looking for bridges */522 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);523 524 if ((header_type & PCI_HEADER_TYPE_MASK) == PCI_HEADER_TYPE_BRIDGE)525 get_pci_parity_status(dev, 1);526}527 528/*529 * PCI Parity polling530 *531 * Function to retrieve the current parity status532 * and decode it533 *534 */535static void edac_pci_dev_parity_test(struct pci_dev *dev)536{537 unsigned long flags;538 u16 status;539 u8 header_type;540 541 /* stop any interrupts until we can acquire the status */542 local_irq_save(flags);543 544 /* read the STATUS register on this device */545 status = get_pci_parity_status(dev, 0);546 547 /* read the device TYPE, looking for bridges */548 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);549 550 local_irq_restore(flags);551 552 edac_dbg(4, "PCI STATUS= 0x%04x %s\n", status, dev_name(&dev->dev));553 554 /* check the status reg for errors on boards NOT marked as broken555 * if broken, we cannot trust any of the status bits556 */557 if (status && !dev->broken_parity_status) {558 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {559 edac_printk(KERN_CRIT, EDAC_PCI,560 "Signaled System Error on %s\n",561 pci_name(dev));562 atomic_inc(&pci_nonparity_count);563 }564 565 if (status & (PCI_STATUS_PARITY)) {566 edac_printk(KERN_CRIT, EDAC_PCI,567 "Master Data Parity Error on %s\n",568 pci_name(dev));569 570 atomic_inc(&pci_parity_count);571 }572 573 if (status & (PCI_STATUS_DETECTED_PARITY)) {574 edac_printk(KERN_CRIT, EDAC_PCI,575 "Detected Parity Error on %s\n",576 pci_name(dev));577 578 atomic_inc(&pci_parity_count);579 }580 }581 582 583 edac_dbg(4, "PCI HEADER TYPE= 0x%02x %s\n",584 header_type, dev_name(&dev->dev));585 586 if ((header_type & PCI_HEADER_TYPE_MASK) == PCI_HEADER_TYPE_BRIDGE) {587 /* On bridges, need to examine secondary status register */588 status = get_pci_parity_status(dev, 1);589 590 edac_dbg(4, "PCI SEC_STATUS= 0x%04x %s\n",591 status, dev_name(&dev->dev));592 593 /* check the secondary status reg for errors,594 * on NOT broken boards595 */596 if (status && !dev->broken_parity_status) {597 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) {598 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "599 "Signaled System Error on %s\n",600 pci_name(dev));601 atomic_inc(&pci_nonparity_count);602 }603 604 if (status & (PCI_STATUS_PARITY)) {605 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "606 "Master Data Parity Error on "607 "%s\n", pci_name(dev));608 609 atomic_inc(&pci_parity_count);610 }611 612 if (status & (PCI_STATUS_DETECTED_PARITY)) {613 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge "614 "Detected Parity Error on %s\n",615 pci_name(dev));616 617 atomic_inc(&pci_parity_count);618 }619 }620 }621}622 623/* reduce some complexity in definition of the iterator */624typedef void (*pci_parity_check_fn_t) (struct pci_dev *dev);625 626/*627 * pci_dev parity list iterator628 *629 * Scan the PCI device list looking for SERRORs, Master Parity ERRORS or630 * Parity ERRORs on primary or secondary devices.631 */632static inline void edac_pci_dev_parity_iterator(pci_parity_check_fn_t fn)633{634 struct pci_dev *dev = NULL;635 636 for_each_pci_dev(dev)637 fn(dev);638}639 640/*641 * edac_pci_do_parity_check642 *643 * performs the actual PCI parity check operation644 */645void edac_pci_do_parity_check(void)646{647 int before_count;648 649 edac_dbg(3, "\n");650 651 /* if policy has PCI check off, leave now */652 if (!check_pci_errors)653 return;654 655 before_count = atomic_read(&pci_parity_count);656 657 /* scan all PCI devices looking for a Parity Error on devices and658 * bridges.659 * The iterator calls pci_get_device() which might sleep, thus660 * we cannot disable interrupts in this scan.661 */662 edac_pci_dev_parity_iterator(edac_pci_dev_parity_test);663 664 /* Only if operator has selected panic on PCI Error */665 if (edac_pci_get_panic_on_pe()) {666 /* If the count is different 'after' from 'before' */667 if (before_count != atomic_read(&pci_parity_count))668 panic("EDAC: PCI Parity Error");669 }670}671 672/*673 * edac_pci_clear_parity_errors674 *675 * function to perform an iteration over the PCI devices676 * and clearn their current status677 */678void edac_pci_clear_parity_errors(void)679{680 /* Clear any PCI bus parity errors that devices initially have logged681 * in their registers.682 */683 edac_pci_dev_parity_iterator(edac_pci_dev_parity_clear);684}685 686/*687 * edac_pci_handle_pe688 *689 * Called to handle a PARITY ERROR event690 */691void edac_pci_handle_pe(struct edac_pci_ctl_info *pci, const char *msg)692{693 694 /* global PE counter incremented by edac_pci_do_parity_check() */695 atomic_inc(&pci->counters.pe_count);696 697 if (edac_pci_get_log_pe())698 edac_pci_printk(pci, KERN_WARNING,699 "Parity Error ctl: %s %d: %s\n",700 pci->ctl_name, pci->pci_idx, msg);701 702 /*703 * poke all PCI devices and see which one is the troublemaker704 * panic() is called if set705 */706 edac_pci_do_parity_check();707}708EXPORT_SYMBOL_GPL(edac_pci_handle_pe);709 710 711/*712 * edac_pci_handle_npe713 *714 * Called to handle a NON-PARITY ERROR event715 */716void edac_pci_handle_npe(struct edac_pci_ctl_info *pci, const char *msg)717{718 719 /* global NPE counter incremented by edac_pci_do_parity_check() */720 atomic_inc(&pci->counters.npe_count);721 722 if (edac_pci_get_log_npe())723 edac_pci_printk(pci, KERN_WARNING,724 "Non-Parity Error ctl: %s %d: %s\n",725 pci->ctl_name, pci->pci_idx, msg);726 727 /*728 * poke all PCI devices and see which one is the troublemaker729 * panic() is called if set730 */731 edac_pci_do_parity_check();732}733EXPORT_SYMBOL_GPL(edac_pci_handle_npe);734 735/*736 * Define the PCI parameter to the module737 */738module_param(check_pci_errors, int, 0644);739MODULE_PARM_DESC(check_pci_errors,740 "Check for PCI bus parity errors: 0=off 1=on");741module_param(edac_pci_panic_on_pe, int, 0644);742MODULE_PARM_DESC(edac_pci_panic_on_pe,743 "Panic on PCI Bus Parity error: 0=off 1=on");744