2 * drivers/pci/pcie/aer/aerdrv_core.c
4 * This file is subject to the terms and conditions of the GNU General Public
5 * License. See the file "COPYING" in the main directory of this archive
8 * This file implements the core part of PCI-Express AER. When an pci-express
9 * error is delivered, an error message will be collected and printed to
10 * console, then, an error recovery procedure will be executed by following
11 * the pci error recovery rules.
13 * Copyright (C) 2006 Intel Corp.
14 * Tom Long Nguyen (tom.l.nguyen@intel.com)
15 * Zhang Yanmin (yanmin.zhang@intel.com)
19 #include <linux/module.h>
20 #include <linux/pci.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
24 #include <linux/suspend.h>
25 #include <linux/delay.h>
29 module_param(forceload
, bool, 0);
31 #define PCI_CFG_SPACE_SIZE (0x100)
32 int pci_find_aer_capability(struct pci_dev
*dev
)
37 /* Check if it's a pci-express device */
38 pos
= pci_find_capability(dev
, PCI_CAP_ID_EXP
);
42 /* Check if it supports pci-express AER */
43 pos
= PCI_CFG_SPACE_SIZE
;
45 if (pci_read_config_dword(dev
, pos
, ®32
))
48 /* some broken boards return ~0 */
49 if (reg32
== 0xffffffff)
52 if (PCI_EXT_CAP_ID(reg32
) == PCI_EXT_CAP_ID_ERR
)
61 int pci_enable_pcie_error_reporting(struct pci_dev
*dev
)
66 pos
= pci_find_capability(dev
, PCI_CAP_ID_EXP
);
70 pci_read_config_word(dev
, pos
+PCI_EXP_DEVCTL
, ®16
);
73 PCI_EXP_DEVCTL_NFERE
|
76 pci_write_config_word(dev
, pos
+PCI_EXP_DEVCTL
,
81 int pci_disable_pcie_error_reporting(struct pci_dev
*dev
)
86 pos
= pci_find_capability(dev
, PCI_CAP_ID_EXP
);
90 pci_read_config_word(dev
, pos
+PCI_EXP_DEVCTL
, ®16
);
91 reg16
= reg16
& ~(PCI_EXP_DEVCTL_CERE
|
92 PCI_EXP_DEVCTL_NFERE
|
95 pci_write_config_word(dev
, pos
+PCI_EXP_DEVCTL
,
100 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev
*dev
)
105 pos
= pci_find_aer_capability(dev
);
109 pci_read_config_dword(dev
, pos
+ PCI_ERR_UNCOR_STATUS
, &status
);
110 pci_read_config_dword(dev
, pos
+ PCI_ERR_UNCOR_SEVER
, &mask
);
111 if (dev
->error_state
== pci_channel_io_normal
)
112 status
&= ~mask
; /* Clear corresponding nonfatal bits */
114 status
&= mask
; /* Clear corresponding fatal bits */
115 pci_write_config_dword(dev
, pos
+ PCI_ERR_UNCOR_STATUS
, status
);
120 int pci_cleanup_aer_correct_error_status(struct pci_dev
*dev
)
125 pos
= pci_find_aer_capability(dev
);
129 pci_read_config_dword(dev
, pos
+ PCI_ERR_COR_STATUS
, &status
);
130 pci_write_config_dword(dev
, pos
+ PCI_ERR_COR_STATUS
, status
);
135 static int find_device_iter(struct device
*device
, void *data
)
138 u16 id
= *(unsigned long *)data
;
139 u8 secondary
, subordinate
, d_bus
= id
>> 8;
141 if (device
->bus
== &pci_bus_type
) {
142 dev
= to_pci_dev(device
);
143 if (id
== ((dev
->bus
->number
<< 8) | dev
->devfn
)) {
147 *(unsigned long*)data
= (unsigned long)device
;
152 * If device is P2P, check if it is an upstream?
154 if (dev
->hdr_type
& PCI_HEADER_TYPE_BRIDGE
) {
155 pci_read_config_byte(dev
, PCI_SECONDARY_BUS
,
157 pci_read_config_byte(dev
, PCI_SUBORDINATE_BUS
,
159 if (d_bus
>= secondary
&& d_bus
<= subordinate
) {
160 *(unsigned long*)data
= (unsigned long)device
;
170 * find_source_device - search through device hierarchy for source device
171 * @parent: pointer to Root Port pci_dev data structure
172 * @id: device ID of agent who sends an error message to this Root Port
174 * Invoked when error is detected at the Root Port.
176 static struct device
* find_source_device(struct pci_dev
*parent
, u16 id
)
178 struct pci_dev
*dev
= parent
;
179 struct device
*device
;
180 unsigned long device_addr
;
183 /* Is Root Port an agent that sends error message? */
184 if (id
== ((dev
->bus
->number
<< 8) | dev
->devfn
))
189 if ((status
= device_for_each_child(&dev
->dev
,
190 &device_addr
, find_device_iter
))) {
191 device
= (struct device
*)device_addr
;
192 dev
= to_pci_dev(device
);
193 if (id
== ((dev
->bus
->number
<< 8) | dev
->devfn
))
201 static void report_error_detected(struct pci_dev
*dev
, void *data
)
203 pci_ers_result_t vote
;
204 struct pci_error_handlers
*err_handler
;
205 struct aer_broadcast_data
*result_data
;
206 result_data
= (struct aer_broadcast_data
*) data
;
208 dev
->error_state
= result_data
->state
;
211 !dev
->driver
->err_handler
||
212 !dev
->driver
->err_handler
->error_detected
) {
213 if (result_data
->state
== pci_channel_io_frozen
&&
214 !(dev
->hdr_type
& PCI_HEADER_TYPE_BRIDGE
)) {
216 * In case of fatal recovery, if one of down-
217 * stream device has no driver. We might be
218 * unable to recover because a later insmod
219 * of a driver for this device is unaware of
222 printk(KERN_DEBUG
"Device ID[%s] has %s\n",
223 dev
->dev
.bus_id
, (dev
->driver
) ?
224 "no AER-aware driver" : "no driver");
229 err_handler
= dev
->driver
->err_handler
;
230 vote
= err_handler
->error_detected(dev
, result_data
->state
);
231 result_data
->result
= merge_result(result_data
->result
, vote
);
235 static void report_mmio_enabled(struct pci_dev
*dev
, void *data
)
237 pci_ers_result_t vote
;
238 struct pci_error_handlers
*err_handler
;
239 struct aer_broadcast_data
*result_data
;
240 result_data
= (struct aer_broadcast_data
*) data
;
243 !dev
->driver
->err_handler
||
244 !dev
->driver
->err_handler
->mmio_enabled
)
247 err_handler
= dev
->driver
->err_handler
;
248 vote
= err_handler
->mmio_enabled(dev
);
249 result_data
->result
= merge_result(result_data
->result
, vote
);
253 static void report_slot_reset(struct pci_dev
*dev
, void *data
)
255 pci_ers_result_t vote
;
256 struct pci_error_handlers
*err_handler
;
257 struct aer_broadcast_data
*result_data
;
258 result_data
= (struct aer_broadcast_data
*) data
;
261 !dev
->driver
->err_handler
||
262 !dev
->driver
->err_handler
->slot_reset
)
265 err_handler
= dev
->driver
->err_handler
;
266 vote
= err_handler
->slot_reset(dev
);
267 result_data
->result
= merge_result(result_data
->result
, vote
);
271 static void report_resume(struct pci_dev
*dev
, void *data
)
273 struct pci_error_handlers
*err_handler
;
275 dev
->error_state
= pci_channel_io_normal
;
278 !dev
->driver
->err_handler
||
279 !dev
->driver
->err_handler
->slot_reset
)
282 err_handler
= dev
->driver
->err_handler
;
283 err_handler
->resume(dev
);
288 * broadcast_error_message - handle message broadcast to downstream drivers
289 * @dev: pointer to from where in a hierarchy message is broadcasted down
290 * @state: error state
291 * @error_mesg: message to print
292 * @cb: callback to be broadcasted
294 * Invoked during error recovery process. Once being invoked, the content
295 * of error severity will be broadcasted to all downstream drivers in a
296 * hierarchy in question.
298 static pci_ers_result_t
broadcast_error_message(struct pci_dev
*dev
,
299 enum pci_channel_state state
,
301 void (*cb
)(struct pci_dev
*, void *))
303 struct aer_broadcast_data result_data
;
305 printk(KERN_DEBUG
"Broadcast %s message\n", error_mesg
);
306 result_data
.state
= state
;
307 if (cb
== report_error_detected
)
308 result_data
.result
= PCI_ERS_RESULT_CAN_RECOVER
;
310 result_data
.result
= PCI_ERS_RESULT_RECOVERED
;
312 if (dev
->hdr_type
& PCI_HEADER_TYPE_BRIDGE
) {
314 * If the error is reported by a bridge, we think this error
315 * is related to the downstream link of the bridge, so we
316 * do error recovery on all subordinates of the bridge instead
317 * of the bridge and clear the error status of the bridge.
319 if (cb
== report_error_detected
)
320 dev
->error_state
= state
;
321 pci_walk_bus(dev
->subordinate
, cb
, &result_data
);
322 if (cb
== report_resume
) {
323 pci_cleanup_aer_uncorrect_error_status(dev
);
324 dev
->error_state
= pci_channel_io_normal
;
329 * If the error is reported by an end point, we think this
330 * error is related to the upstream link of the end point.
332 pci_walk_bus(dev
->bus
, cb
, &result_data
);
335 return result_data
.result
;
338 struct find_aer_service_data
{
339 struct pcie_port_service_driver
*aer_driver
;
343 static int find_aer_service_iter(struct device
*device
, void *data
)
345 struct device_driver
*driver
;
346 struct pcie_port_service_driver
*service_driver
;
347 struct pcie_device
*pcie_dev
;
348 struct find_aer_service_data
*result
;
350 result
= (struct find_aer_service_data
*) data
;
352 if (device
->bus
== &pcie_port_bus_type
) {
353 pcie_dev
= to_pcie_device(device
);
354 if (pcie_dev
->id
.port_type
== PCIE_SW_DOWNSTREAM_PORT
)
355 result
->is_downstream
= 1;
357 driver
= device
->driver
;
359 service_driver
= to_service_driver(driver
);
360 if (service_driver
->id_table
->service_type
==
361 PCIE_PORT_SERVICE_AER
) {
362 result
->aer_driver
= service_driver
;
371 static void find_aer_service(struct pci_dev
*dev
,
372 struct find_aer_service_data
*data
)
375 retval
= device_for_each_child(&dev
->dev
, data
, find_aer_service_iter
);
378 static pci_ers_result_t
reset_link(struct pcie_device
*aerdev
,
381 struct pci_dev
*udev
;
382 pci_ers_result_t status
;
383 struct find_aer_service_data data
;
385 if (dev
->hdr_type
& PCI_HEADER_TYPE_BRIDGE
)
388 udev
= dev
->bus
->self
;
390 data
.is_downstream
= 0;
391 data
.aer_driver
= NULL
;
392 find_aer_service(udev
, &data
);
395 * Use the aer driver of the error agent firstly.
396 * If it hasn't the aer driver, use the root port's
398 if (!data
.aer_driver
|| !data
.aer_driver
->reset_link
) {
399 if (data
.is_downstream
&&
400 aerdev
->device
.driver
&&
401 to_service_driver(aerdev
->device
.driver
)->reset_link
) {
403 to_service_driver(aerdev
->device
.driver
);
405 printk(KERN_DEBUG
"No link-reset support to Device ID"
408 return PCI_ERS_RESULT_DISCONNECT
;
412 status
= data
.aer_driver
->reset_link(udev
);
413 if (status
!= PCI_ERS_RESULT_RECOVERED
) {
414 printk(KERN_DEBUG
"Link reset at upstream Device ID"
417 return PCI_ERS_RESULT_DISCONNECT
;
424 * do_recovery - handle nonfatal/fatal error recovery process
425 * @aerdev: pointer to a pcie_device data structure of root port
426 * @dev: pointer to a pci_dev data structure of agent detecting an error
427 * @severity: error severity type
429 * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
430 * error detected message to all downstream drivers within a hierarchy in
431 * question and return the returned code.
433 static pci_ers_result_t
do_recovery(struct pcie_device
*aerdev
,
437 pci_ers_result_t status
, result
= PCI_ERS_RESULT_RECOVERED
;
438 enum pci_channel_state state
;
440 if (severity
== AER_FATAL
)
441 state
= pci_channel_io_frozen
;
443 state
= pci_channel_io_normal
;
445 status
= broadcast_error_message(dev
,
448 report_error_detected
);
450 if (severity
== AER_FATAL
) {
451 result
= reset_link(aerdev
, dev
);
452 if (result
!= PCI_ERS_RESULT_RECOVERED
) {
453 /* TODO: Should panic here? */
458 if (status
== PCI_ERS_RESULT_CAN_RECOVER
)
459 status
= broadcast_error_message(dev
,
462 report_mmio_enabled
);
464 if (status
== PCI_ERS_RESULT_NEED_RESET
) {
466 * TODO: Should call platform-specific
467 * functions to reset slot before calling
468 * drivers' slot_reset callbacks?
470 status
= broadcast_error_message(dev
,
476 if (status
== PCI_ERS_RESULT_RECOVERED
)
477 broadcast_error_message(dev
,
486 * handle_error_source - handle logging error into an event log
487 * @aerdev: pointer to pcie_device data structure of the root port
488 * @dev: pointer to pci_dev data structure of error source device
489 * @info: comprehensive error information
491 * Invoked when an error being detected by Root Port.
493 static void handle_error_source(struct pcie_device
* aerdev
,
495 struct aer_err_info info
)
497 pci_ers_result_t status
= 0;
500 if (info
.severity
== AER_CORRECTABLE
) {
502 * Correctable error does not need software intevention.
503 * No need to go through error recovery process.
505 pos
= pci_find_aer_capability(dev
);
507 pci_write_config_dword(dev
, pos
+ PCI_ERR_COR_STATUS
,
510 status
= do_recovery(aerdev
, dev
, info
.severity
);
511 if (status
== PCI_ERS_RESULT_RECOVERED
) {
512 printk(KERN_DEBUG
"AER driver successfully recovered\n");
514 /* TODO: Should kernel panic here? */
515 printk(KERN_DEBUG
"AER driver didn't recover\n");
521 * aer_enable_rootport - enable Root Port's interrupts when receiving messages
522 * @rpc: pointer to a Root Port data structure
524 * Invoked when PCIE bus loads AER service driver.
526 void aer_enable_rootport(struct aer_rpc
*rpc
)
528 struct pci_dev
*pdev
= rpc
->rpd
->port
;
533 pos
= pci_find_capability(pdev
, PCI_CAP_ID_EXP
);
534 /* Clear PCIE Capability's Device Status */
535 pci_read_config_word(pdev
, pos
+PCI_EXP_DEVSTA
, ®16
);
536 pci_write_config_word(pdev
, pos
+PCI_EXP_DEVSTA
, reg16
);
538 /* Disable system error generation in response to error messages */
539 pci_read_config_word(pdev
, pos
+ PCI_EXP_RTCTL
, ®16
);
540 reg16
&= ~(SYSTEM_ERROR_INTR_ON_MESG_MASK
);
541 pci_write_config_word(pdev
, pos
+ PCI_EXP_RTCTL
, reg16
);
543 aer_pos
= pci_find_aer_capability(pdev
);
544 /* Clear error status */
545 pci_read_config_dword(pdev
, aer_pos
+ PCI_ERR_ROOT_STATUS
, ®32
);
546 pci_write_config_dword(pdev
, aer_pos
+ PCI_ERR_ROOT_STATUS
, reg32
);
547 pci_read_config_dword(pdev
, aer_pos
+ PCI_ERR_COR_STATUS
, ®32
);
548 pci_write_config_dword(pdev
, aer_pos
+ PCI_ERR_COR_STATUS
, reg32
);
549 pci_read_config_dword(pdev
, aer_pos
+ PCI_ERR_UNCOR_STATUS
, ®32
);
550 pci_write_config_dword(pdev
, aer_pos
+ PCI_ERR_UNCOR_STATUS
, reg32
);
552 /* Enable Root Port device reporting error itself */
553 pci_read_config_word(pdev
, pos
+PCI_EXP_DEVCTL
, ®16
);
555 PCI_EXP_DEVCTL_CERE
|
556 PCI_EXP_DEVCTL_NFERE
|
557 PCI_EXP_DEVCTL_FERE
|
559 pci_write_config_word(pdev
, pos
+PCI_EXP_DEVCTL
,
562 /* Enable Root Port's interrupt in response to error messages */
563 pci_write_config_dword(pdev
,
564 aer_pos
+ PCI_ERR_ROOT_COMMAND
,
565 ROOT_PORT_INTR_ON_MESG_MASK
);
569 * disable_root_aer - disable Root Port's interrupts when receiving messages
570 * @rpc: pointer to a Root Port data structure
572 * Invoked when PCIE bus unloads AER service driver.
574 static void disable_root_aer(struct aer_rpc
*rpc
)
576 struct pci_dev
*pdev
= rpc
->rpd
->port
;
580 pos
= pci_find_aer_capability(pdev
);
581 /* Disable Root's interrupt in response to error messages */
582 pci_write_config_dword(pdev
, pos
+ PCI_ERR_ROOT_COMMAND
, 0);
584 /* Clear Root's error status reg */
585 pci_read_config_dword(pdev
, pos
+ PCI_ERR_ROOT_STATUS
, ®32
);
586 pci_write_config_dword(pdev
, pos
+ PCI_ERR_ROOT_STATUS
, reg32
);
590 * get_e_source - retrieve an error source
591 * @rpc: pointer to the root port which holds an error
593 * Invoked by DPC handler to consume an error.
595 static struct aer_err_source
* get_e_source(struct aer_rpc
*rpc
)
597 struct aer_err_source
*e_source
;
600 /* Lock access to Root error producer/consumer index */
601 spin_lock_irqsave(&rpc
->e_lock
, flags
);
602 if (rpc
->prod_idx
== rpc
->cons_idx
) {
603 spin_unlock_irqrestore(&rpc
->e_lock
, flags
);
606 e_source
= &rpc
->e_sources
[rpc
->cons_idx
];
608 if (rpc
->cons_idx
== AER_ERROR_SOURCES_MAX
)
610 spin_unlock_irqrestore(&rpc
->e_lock
, flags
);
615 static int get_device_error_info(struct pci_dev
*dev
, struct aer_err_info
*info
)
619 pos
= pci_find_aer_capability(dev
);
621 /* The device might not support AER */
625 if (info
->severity
== AER_CORRECTABLE
) {
626 pci_read_config_dword(dev
, pos
+ PCI_ERR_COR_STATUS
,
628 if (!(info
->status
& ERR_CORRECTABLE_ERROR_MASK
))
629 return AER_UNSUCCESS
;
630 } else if (dev
->hdr_type
& PCI_HEADER_TYPE_BRIDGE
||
631 info
->severity
== AER_NONFATAL
) {
633 /* Link is still healthy for IO reads */
634 pci_read_config_dword(dev
, pos
+ PCI_ERR_UNCOR_STATUS
,
636 if (!(info
->status
& ERR_UNCORRECTABLE_ERROR_MASK
))
637 return AER_UNSUCCESS
;
639 if (info
->status
& AER_LOG_TLP_MASKS
) {
640 info
->flags
|= AER_TLP_HEADER_VALID_FLAG
;
641 pci_read_config_dword(dev
,
642 pos
+ PCI_ERR_HEADER_LOG
, &info
->tlp
.dw0
);
643 pci_read_config_dword(dev
,
644 pos
+ PCI_ERR_HEADER_LOG
+ 4, &info
->tlp
.dw1
);
645 pci_read_config_dword(dev
,
646 pos
+ PCI_ERR_HEADER_LOG
+ 8, &info
->tlp
.dw2
);
647 pci_read_config_dword(dev
,
648 pos
+ PCI_ERR_HEADER_LOG
+ 12, &info
->tlp
.dw3
);
656 * aer_isr_one_error - consume an error detected by root port
657 * @p_device: pointer to error root port service device
658 * @e_src: pointer to an error source
660 static void aer_isr_one_error(struct pcie_device
*p_device
,
661 struct aer_err_source
*e_src
)
663 struct device
*s_device
;
664 struct aer_err_info e_info
= {0, 0, 0,};
669 * There is a possibility that both correctable error and
670 * uncorrectable error being logged. Report correctable error first.
672 for (i
= 1; i
& ROOT_ERR_STATUS_MASKS
; i
<<= 2) {
675 if (!(e_src
->status
& i
))
678 /* Init comprehensive error information */
679 if (i
& PCI_ERR_ROOT_COR_RCV
) {
680 id
= ERR_COR_ID(e_src
->id
);
681 e_info
.severity
= AER_CORRECTABLE
;
683 id
= ERR_UNCOR_ID(e_src
->id
);
684 e_info
.severity
= ((e_src
->status
>> 6) & 1);
687 (PCI_ERR_ROOT_MULTI_COR_RCV
|
688 PCI_ERR_ROOT_MULTI_UNCOR_RCV
))
689 e_info
.flags
|= AER_MULTI_ERROR_VALID_FLAG
;
690 if (!(s_device
= find_source_device(p_device
->port
, id
))) {
691 printk(KERN_DEBUG
"%s->can't find device of ID%04x\n",
695 if (get_device_error_info(to_pci_dev(s_device
), &e_info
) ==
697 aer_print_error(to_pci_dev(s_device
), &e_info
);
698 handle_error_source(p_device
,
699 to_pci_dev(s_device
),
706 * aer_isr - consume errors detected by root port
707 * @work: definition of this work item
709 * Invoked, as DPC, when root port records new detected error
711 void aer_isr(struct work_struct
*work
)
713 struct aer_rpc
*rpc
= container_of(work
, struct aer_rpc
, dpc_handler
);
714 struct pcie_device
*p_device
= rpc
->rpd
;
715 struct aer_err_source
*e_src
;
717 mutex_lock(&rpc
->rpc_mutex
);
718 e_src
= get_e_source(rpc
);
720 aer_isr_one_error(p_device
, e_src
);
721 e_src
= get_e_source(rpc
);
723 mutex_unlock(&rpc
->rpc_mutex
);
725 wake_up(&rpc
->wait_release
);
729 * aer_delete_rootport - disable root port aer and delete service data
730 * @rpc: pointer to a root port device being deleted
732 * Invoked when AER service unloaded on a specific Root Port
734 void aer_delete_rootport(struct aer_rpc
*rpc
)
736 /* Disable root port AER itself */
737 disable_root_aer(rpc
);
743 * aer_init - provide AER initialization
744 * @dev: pointer to AER pcie device
746 * Invoked when AER service driver is loaded.
748 int aer_init(struct pcie_device
*dev
)
750 if (aer_osc_setup(dev
) && !forceload
)
756 EXPORT_SYMBOL_GPL(pci_find_aer_capability
);
757 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting
);
758 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting
);
759 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status
);
760 EXPORT_SYMBOL_GPL(pci_cleanup_aer_correct_error_status
);