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 const struct bus_type
*edac_subsys
;
342 struct device
*dev_root
;
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();
355 /* Bump the reference count on this module to ensure the
356 * modules isn't unloaded until we deconstruct the top
357 * level main kobj for EDAC PCI
359 if (!try_module_get(THIS_MODULE
)) {
360 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 dev_root
= bus_get_dev_root(edac_subsys
);
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
);
380 edac_dbg(1, "Failed to register '.../edac/pci'\n");
381 goto kobject_init_and_add_fail
;
384 /* At this point, to 'release' the top level kobject
385 * for EDAC PCI, then edac_pci_main_kobj_teardown()
386 * must be used, for resources to be cleaned up properly
388 kobject_uevent(edac_pci_top_main_kobj
, KOBJ_ADD
);
389 edac_dbg(1, "Registered '.../edac/pci' kobject\n");
393 /* Error unwind statck */
394 kobject_init_and_add_fail
:
395 kobject_put(edac_pci_top_main_kobj
);
398 module_put(THIS_MODULE
);
400 decrement_count_fail
:
401 /* if are on this error exit, nothing to tear down */
402 atomic_dec(&edac_pci_sysfs_refcount
);
408 * edac_pci_main_kobj_teardown()
410 * if no longer linked (needed) remove the top level EDAC PCI
411 * kobject with its controls and attributes
413 static void edac_pci_main_kobj_teardown(void)
417 /* Decrement the count and only if no more controller instances
418 * are connected perform the unregisteration of the top level
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
);
427 int edac_pci_create_sysfs(struct edac_pci_ctl_info
*pci
)
430 struct kobject
*edac_kobj
= &pci
->kobj
;
432 edac_dbg(0, "idx=%d\n", pci
->pci_idx
);
434 /* create the top main EDAC PCI kobject, IF needed */
435 err
= edac_pci_main_kobj_setup();
439 /* Create this instance's kobject under the MAIN kobject */
440 err
= edac_pci_create_instance_kobj(pci
, pci
->pci_idx
);
442 goto unregister_cleanup
;
444 err
= sysfs_create_link(edac_kobj
, &pci
->dev
->kobj
, EDAC_PCI_SYMLINK
);
446 edac_dbg(0, "sysfs_create_link() returned err= %d\n", err
);
452 /* Error unwind stack */
454 edac_pci_unregister_sysfs_instance_kobj(pci
);
457 edac_pci_main_kobj_teardown();
462 void edac_pci_remove_sysfs(struct edac_pci_ctl_info
*pci
)
464 edac_dbg(0, "index=%d\n", pci
->pci_idx
);
466 /* Remove the symlink */
467 sysfs_remove_link(&pci
->kobj
, EDAC_PCI_SYMLINK
);
469 /* remove this PCI instance's sysfs entries */
470 edac_pci_unregister_sysfs_instance_kobj(pci
);
472 /* Call the main unregister function, which will determine
473 * if this 'pci' is the last instance.
474 * If it is, the main kobject will be unregistered as a result
476 edac_dbg(0, "calling edac_pci_main_kobj_teardown()\n");
477 edac_pci_main_kobj_teardown();
480 /************************ PCI error handling *************************/
481 static u16
get_pci_parity_status(struct pci_dev
*dev
, int secondary
)
486 where
= secondary
? PCI_SEC_STATUS
: PCI_STATUS
;
487 pci_read_config_word(dev
, where
, &status
);
489 /* If we get back 0xFFFF then we must suspect that the card has been
490 * pulled but the Linux PCI layer has not yet finished cleaning up.
491 * We don't want to report on such devices
494 if (status
== 0xFFFF) {
497 pci_read_config_dword(dev
, 0, &sanity
);
499 if (sanity
== 0xFFFFFFFF)
503 status
&= PCI_STATUS_DETECTED_PARITY
| PCI_STATUS_SIG_SYSTEM_ERROR
|
507 /* reset only the bits we are interested in */
508 pci_write_config_word(dev
, where
, status
);
514 /* Clear any PCI parity errors logged by this device. */
515 static void edac_pci_dev_parity_clear(struct pci_dev
*dev
)
519 get_pci_parity_status(dev
, 0);
521 /* read the device TYPE, looking for bridges */
522 pci_read_config_byte(dev
, PCI_HEADER_TYPE
, &header_type
);
524 if ((header_type
& PCI_HEADER_TYPE_MASK
) == PCI_HEADER_TYPE_BRIDGE
)
525 get_pci_parity_status(dev
, 1);
531 * Function to retrieve the current parity status
535 static void edac_pci_dev_parity_test(struct pci_dev
*dev
)
541 /* stop any interrupts until we can acquire the status */
542 local_irq_save(flags
);
544 /* read the STATUS register on this device */
545 status
= get_pci_parity_status(dev
, 0);
547 /* read the device TYPE, looking for bridges */
548 pci_read_config_byte(dev
, PCI_HEADER_TYPE
, &header_type
);
550 local_irq_restore(flags
);
552 edac_dbg(4, "PCI STATUS= 0x%04x %s\n", status
, dev_name(&dev
->dev
));
554 /* check the status reg for errors on boards NOT marked as broken
555 * if broken, we cannot trust any of the status bits
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",
562 atomic_inc(&pci_nonparity_count
);
565 if (status
& (PCI_STATUS_PARITY
)) {
566 edac_printk(KERN_CRIT
, EDAC_PCI
,
567 "Master Data Parity Error on %s\n",
570 atomic_inc(&pci_parity_count
);
573 if (status
& (PCI_STATUS_DETECTED_PARITY
)) {
574 edac_printk(KERN_CRIT
, EDAC_PCI
,
575 "Detected Parity Error on %s\n",
578 atomic_inc(&pci_parity_count
);
583 edac_dbg(4, "PCI HEADER TYPE= 0x%02x %s\n",
584 header_type
, dev_name(&dev
->dev
));
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);
590 edac_dbg(4, "PCI SEC_STATUS= 0x%04x %s\n",
591 status
, dev_name(&dev
->dev
));
593 /* check the secondary status reg for errors,
594 * on NOT broken boards
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",
601 atomic_inc(&pci_nonparity_count
);
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
));
609 atomic_inc(&pci_parity_count
);
612 if (status
& (PCI_STATUS_DETECTED_PARITY
)) {
613 edac_printk(KERN_CRIT
, EDAC_PCI
, "Bridge "
614 "Detected Parity Error on %s\n",
617 atomic_inc(&pci_parity_count
);
623 /* reduce some complexity in definition of the iterator */
624 typedef void (*pci_parity_check_fn_t
) (struct pci_dev
*dev
);
627 * pci_dev parity list iterator
629 * Scan the PCI device list looking for SERRORs, Master Parity ERRORS or
630 * Parity ERRORs on primary or secondary devices.
632 static inline void edac_pci_dev_parity_iterator(pci_parity_check_fn_t fn
)
634 struct pci_dev
*dev
= NULL
;
636 for_each_pci_dev(dev
)
641 * edac_pci_do_parity_check
643 * performs the actual PCI parity check operation
645 void edac_pci_do_parity_check(void)
651 /* if policy has PCI check off, leave now */
652 if (!check_pci_errors
)
655 before_count
= atomic_read(&pci_parity_count
);
657 /* scan all PCI devices looking for a Parity Error on devices and
659 * The iterator calls pci_get_device() which might sleep, thus
660 * we cannot disable interrupts in this scan.
662 edac_pci_dev_parity_iterator(edac_pci_dev_parity_test
);
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");
673 * edac_pci_clear_parity_errors
675 * function to perform an iteration over the PCI devices
676 * and clearn their current status
678 void edac_pci_clear_parity_errors(void)
680 /* Clear any PCI bus parity errors that devices initially have logged
681 * in their registers.
683 edac_pci_dev_parity_iterator(edac_pci_dev_parity_clear
);
689 * Called to handle a PARITY ERROR event
691 void edac_pci_handle_pe(struct edac_pci_ctl_info
*pci
, const char *msg
)
694 /* global PE counter incremented by edac_pci_do_parity_check() */
695 atomic_inc(&pci
->counters
.pe_count
);
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
);
703 * poke all PCI devices and see which one is the troublemaker
704 * panic() is called if set
706 edac_pci_do_parity_check();
708 EXPORT_SYMBOL_GPL(edac_pci_handle_pe
);
712 * edac_pci_handle_npe
714 * Called to handle a NON-PARITY ERROR event
716 void edac_pci_handle_npe(struct edac_pci_ctl_info
*pci
, const char *msg
)
719 /* global NPE counter incremented by edac_pci_do_parity_check() */
720 atomic_inc(&pci
->counters
.npe_count
);
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
);
728 * poke all PCI devices and see which one is the troublemaker
729 * panic() is called if set
731 edac_pci_do_parity_check();
733 EXPORT_SYMBOL_GPL(edac_pci_handle_npe
);
736 * Define the PCI parameter to the module
738 module_param(check_pci_errors
, int, 0644);
739 MODULE_PARM_DESC(check_pci_errors
,
740 "Check for PCI bus parity errors: 0=off 1=on");
741 module_param(edac_pci_panic_on_pe
, int, 0644);
742 MODULE_PARM_DESC(edac_pci_panic_on_pe
,
743 "Panic on PCI Bus Parity error: 0=off 1=on");