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 static int nosourceid;
30 module_param(forceload, bool, 0);
31 module_param(nosourceid, bool, 0);
33 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
38 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
42 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
46 pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, ®16);
49 PCI_EXP_DEVCTL_NFERE |
52 pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
56 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
58 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
63 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
67 pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, ®16);
68 reg16 = reg16 & ~(PCI_EXP_DEVCTL_CERE |
69 PCI_EXP_DEVCTL_NFERE |
72 pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
76 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
78 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
83 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
87 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
88 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
89 if (dev->error_state == pci_channel_io_normal)
90 status &= ~mask; /* Clear corresponding nonfatal bits */
92 status &= mask; /* Clear corresponding fatal bits */
93 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
97 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
100 int pci_cleanup_aer_correct_error_status(struct pci_dev *dev)
105 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
109 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
110 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
116 static int set_device_error_reporting(struct pci_dev *dev, void *data)
118 bool enable = *((bool *)data);
120 if (dev->pcie_type == PCIE_RC_PORT ||
121 dev->pcie_type == PCIE_SW_UPSTREAM_PORT ||
122 dev->pcie_type == PCIE_SW_DOWNSTREAM_PORT) {
124 pci_enable_pcie_error_reporting(dev);
126 pci_disable_pcie_error_reporting(dev);
130 pcie_set_ecrc_checking(dev);
136 * set_downstream_devices_error_reporting - enable/disable the error reporting bits on the root port and its downstream ports.
137 * @dev: pointer to root port's pci_dev data structure
138 * @enable: true = enable error reporting, false = disable error reporting.
140 static void set_downstream_devices_error_reporting(struct pci_dev *dev,
143 set_device_error_reporting(dev, &enable);
145 if (!dev->subordinate)
147 pci_walk_bus(dev->subordinate, set_device_error_reporting, &enable);
150 static inline int compare_device_id(struct pci_dev *dev,
151 struct aer_err_info *e_info)
153 if (e_info->id == ((dev->bus->number << 8) | dev->devfn)) {
163 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
165 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
166 e_info->dev[e_info->error_dev_num] = dev;
167 e_info->error_dev_num++;
175 #define PCI_BUS(x) (((x) >> 8) & 0xff)
177 static int find_device_iter(struct pci_dev *dev, void *data)
184 struct aer_err_info *e_info = (struct aer_err_info *)data;
187 * When bus id is equal to 0, it might be a bad id
188 * reported by root port.
190 if (!nosourceid && (PCI_BUS(e_info->id) != 0)) {
191 result = compare_device_id(dev, e_info);
193 add_error_device(e_info, dev);
196 * If there is no multiple error, we stop
197 * or continue based on the id comparing.
199 if (!(e_info->flags & AER_MULTI_ERROR_VALID_FLAG))
203 * If there are multiple errors and id does match,
204 * We need continue to search other devices under
205 * the root port. Return 0 means that.
214 * 2) bus id is equal to 0. Some ports might lose the bus
215 * id of error source id;
216 * 3) There are multiple errors and prior id comparing fails;
217 * We check AER status registers to find the initial reporter.
219 if (atomic_read(&dev->enable_cnt) == 0)
221 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
224 /* Check if AER is enabled */
225 pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, ®16);
227 PCI_EXP_DEVCTL_CERE |
228 PCI_EXP_DEVCTL_NFERE |
229 PCI_EXP_DEVCTL_FERE |
230 PCI_EXP_DEVCTL_URRE)))
232 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
238 if (e_info->severity == AER_CORRECTABLE) {
239 pci_read_config_dword(dev,
240 pos + PCI_ERR_COR_STATUS,
242 pci_read_config_dword(dev,
243 pos + PCI_ERR_COR_MASK,
245 if (status & ERR_CORRECTABLE_ERROR_MASK & ~mask) {
246 add_error_device(e_info, dev);
250 pci_read_config_dword(dev,
251 pos + PCI_ERR_UNCOR_STATUS,
253 pci_read_config_dword(dev,
254 pos + PCI_ERR_UNCOR_MASK,
256 if (status & ERR_UNCORRECTABLE_ERROR_MASK & ~mask) {
257 add_error_device(e_info, dev);
265 if (e_info->flags & AER_MULTI_ERROR_VALID_FLAG)
272 * find_source_device - search through device hierarchy for source device
273 * @parent: pointer to Root Port pci_dev data structure
274 * @err_info: including detailed error information such like id
276 * Invoked when error is detected at the Root Port.
278 static void find_source_device(struct pci_dev *parent,
279 struct aer_err_info *e_info)
281 struct pci_dev *dev = parent;
284 /* Is Root Port an agent that sends error message? */
285 result = find_device_iter(dev, e_info);
289 pci_walk_bus(parent->subordinate, find_device_iter, e_info);
292 static int report_error_detected(struct pci_dev *dev, void *data)
294 pci_ers_result_t vote;
295 struct pci_error_handlers *err_handler;
296 struct aer_broadcast_data *result_data;
297 result_data = (struct aer_broadcast_data *) data;
299 dev->error_state = result_data->state;
302 !dev->driver->err_handler ||
303 !dev->driver->err_handler->error_detected) {
304 if (result_data->state == pci_channel_io_frozen &&
305 !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
307 * In case of fatal recovery, if one of down-
308 * stream device has no driver. We might be
309 * unable to recover because a later insmod
310 * of a driver for this device is unaware of
313 dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
315 "no AER-aware driver" : "no driver");
320 err_handler = dev->driver->err_handler;
321 vote = err_handler->error_detected(dev, result_data->state);
322 result_data->result = merge_result(result_data->result, vote);
326 static int report_mmio_enabled(struct pci_dev *dev, void *data)
328 pci_ers_result_t vote;
329 struct pci_error_handlers *err_handler;
330 struct aer_broadcast_data *result_data;
331 result_data = (struct aer_broadcast_data *) data;
334 !dev->driver->err_handler ||
335 !dev->driver->err_handler->mmio_enabled)
338 err_handler = dev->driver->err_handler;
339 vote = err_handler->mmio_enabled(dev);
340 result_data->result = merge_result(result_data->result, vote);
344 static int report_slot_reset(struct pci_dev *dev, void *data)
346 pci_ers_result_t vote;
347 struct pci_error_handlers *err_handler;
348 struct aer_broadcast_data *result_data;
349 result_data = (struct aer_broadcast_data *) data;
352 !dev->driver->err_handler ||
353 !dev->driver->err_handler->slot_reset)
356 err_handler = dev->driver->err_handler;
357 vote = err_handler->slot_reset(dev);
358 result_data->result = merge_result(result_data->result, vote);
362 static int report_resume(struct pci_dev *dev, void *data)
364 struct pci_error_handlers *err_handler;
366 dev->error_state = pci_channel_io_normal;
369 !dev->driver->err_handler ||
370 !dev->driver->err_handler->resume)
373 err_handler = dev->driver->err_handler;
374 err_handler->resume(dev);
379 * broadcast_error_message - handle message broadcast to downstream drivers
380 * @dev: pointer to from where in a hierarchy message is broadcasted down
381 * @state: error state
382 * @error_mesg: message to print
383 * @cb: callback to be broadcasted
385 * Invoked during error recovery process. Once being invoked, the content
386 * of error severity will be broadcasted to all downstream drivers in a
387 * hierarchy in question.
389 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
390 enum pci_channel_state state,
392 int (*cb)(struct pci_dev *, void *))
394 struct aer_broadcast_data result_data;
396 dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
397 result_data.state = state;
398 if (cb == report_error_detected)
399 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
401 result_data.result = PCI_ERS_RESULT_RECOVERED;
403 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
405 * If the error is reported by a bridge, we think this error
406 * is related to the downstream link of the bridge, so we
407 * do error recovery on all subordinates of the bridge instead
408 * of the bridge and clear the error status of the bridge.
410 if (cb == report_error_detected)
411 dev->error_state = state;
412 pci_walk_bus(dev->subordinate, cb, &result_data);
413 if (cb == report_resume) {
414 pci_cleanup_aer_uncorrect_error_status(dev);
415 dev->error_state = pci_channel_io_normal;
419 * If the error is reported by an end point, we think this
420 * error is related to the upstream link of the end point.
422 pci_walk_bus(dev->bus, cb, &result_data);
425 return result_data.result;
428 struct find_aer_service_data {
429 struct pcie_port_service_driver *aer_driver;
433 static int find_aer_service_iter(struct device *device, void *data)
435 struct device_driver *driver;
436 struct pcie_port_service_driver *service_driver;
437 struct find_aer_service_data *result;
439 result = (struct find_aer_service_data *) data;
441 if (device->bus == &pcie_port_bus_type) {
442 struct pcie_port_data *port_data;
444 port_data = pci_get_drvdata(to_pcie_device(device)->port);
445 if (port_data->port_type == PCIE_SW_DOWNSTREAM_PORT)
446 result->is_downstream = 1;
448 driver = device->driver;
450 service_driver = to_service_driver(driver);
451 if (service_driver->service == PCIE_PORT_SERVICE_AER) {
452 result->aer_driver = service_driver;
461 static void find_aer_service(struct pci_dev *dev,
462 struct find_aer_service_data *data)
465 retval = device_for_each_child(&dev->dev, data, find_aer_service_iter);
468 static pci_ers_result_t reset_link(struct pcie_device *aerdev,
471 struct pci_dev *udev;
472 pci_ers_result_t status;
473 struct find_aer_service_data data;
475 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)
478 udev = dev->bus->self;
480 data.is_downstream = 0;
481 data.aer_driver = NULL;
482 find_aer_service(udev, &data);
485 * Use the aer driver of the error agent firstly.
486 * If it hasn't the aer driver, use the root port's
488 if (!data.aer_driver || !data.aer_driver->reset_link) {
489 if (data.is_downstream &&
490 aerdev->device.driver &&
491 to_service_driver(aerdev->device.driver)->reset_link) {
493 to_service_driver(aerdev->device.driver);
495 dev_printk(KERN_DEBUG, &dev->dev, "no link-reset "
497 return PCI_ERS_RESULT_DISCONNECT;
501 status = data.aer_driver->reset_link(udev);
502 if (status != PCI_ERS_RESULT_RECOVERED) {
503 dev_printk(KERN_DEBUG, &dev->dev, "link reset at upstream "
504 "device %s failed\n", pci_name(udev));
505 return PCI_ERS_RESULT_DISCONNECT;
512 * do_recovery - handle nonfatal/fatal error recovery process
513 * @aerdev: pointer to a pcie_device data structure of root port
514 * @dev: pointer to a pci_dev data structure of agent detecting an error
515 * @severity: error severity type
517 * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
518 * error detected message to all downstream drivers within a hierarchy in
519 * question and return the returned code.
521 static pci_ers_result_t do_recovery(struct pcie_device *aerdev,
525 pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
526 enum pci_channel_state state;
528 if (severity == AER_FATAL)
529 state = pci_channel_io_frozen;
531 state = pci_channel_io_normal;
533 status = broadcast_error_message(dev,
536 report_error_detected);
538 if (severity == AER_FATAL) {
539 result = reset_link(aerdev, dev);
540 if (result != PCI_ERS_RESULT_RECOVERED) {
541 /* TODO: Should panic here? */
546 if (status == PCI_ERS_RESULT_CAN_RECOVER)
547 status = broadcast_error_message(dev,
550 report_mmio_enabled);
552 if (status == PCI_ERS_RESULT_NEED_RESET) {
554 * TODO: Should call platform-specific
555 * functions to reset slot before calling
556 * drivers' slot_reset callbacks?
558 status = broadcast_error_message(dev,
564 if (status == PCI_ERS_RESULT_RECOVERED)
565 broadcast_error_message(dev,
574 * handle_error_source - handle logging error into an event log
575 * @aerdev: pointer to pcie_device data structure of the root port
576 * @dev: pointer to pci_dev data structure of error source device
577 * @info: comprehensive error information
579 * Invoked when an error being detected by Root Port.
581 static void handle_error_source(struct pcie_device *aerdev,
583 struct aer_err_info *info)
585 pci_ers_result_t status = 0;
588 if (info->severity == AER_CORRECTABLE) {
590 * Correctable error does not need software intevention.
591 * No need to go through error recovery process.
593 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
595 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
598 status = do_recovery(aerdev, dev, info->severity);
599 if (status == PCI_ERS_RESULT_RECOVERED) {
600 dev_printk(KERN_DEBUG, &dev->dev, "AER driver "
601 "successfully recovered\n");
603 /* TODO: Should kernel panic here? */
604 dev_printk(KERN_DEBUG, &dev->dev, "AER driver didn't "
611 * aer_enable_rootport - enable Root Port's interrupts when receiving messages
612 * @rpc: pointer to a Root Port data structure
614 * Invoked when PCIE bus loads AER service driver.
616 void aer_enable_rootport(struct aer_rpc *rpc)
618 struct pci_dev *pdev = rpc->rpd->port;
623 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
624 /* Clear PCIE Capability's Device Status */
625 pci_read_config_word(pdev, pos+PCI_EXP_DEVSTA, ®16);
626 pci_write_config_word(pdev, pos+PCI_EXP_DEVSTA, reg16);
628 /* Disable system error generation in response to error messages */
629 pci_read_config_word(pdev, pos + PCI_EXP_RTCTL, ®16);
630 reg16 &= ~(SYSTEM_ERROR_INTR_ON_MESG_MASK);
631 pci_write_config_word(pdev, pos + PCI_EXP_RTCTL, reg16);
633 aer_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
634 /* Clear error status */
635 pci_read_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, ®32);
636 pci_write_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, reg32);
637 pci_read_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, ®32);
638 pci_write_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, reg32);
639 pci_read_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, ®32);
640 pci_write_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, reg32);
643 * Enable error reporting for the root port device and downstream port
646 set_downstream_devices_error_reporting(pdev, true);
648 /* Enable Root Port's interrupt in response to error messages */
649 pci_write_config_dword(pdev,
650 aer_pos + PCI_ERR_ROOT_COMMAND,
651 ROOT_PORT_INTR_ON_MESG_MASK);
655 * disable_root_aer - disable Root Port's interrupts when receiving messages
656 * @rpc: pointer to a Root Port data structure
658 * Invoked when PCIE bus unloads AER service driver.
660 static void disable_root_aer(struct aer_rpc *rpc)
662 struct pci_dev *pdev = rpc->rpd->port;
667 * Disable error reporting for the root port device and downstream port
670 set_downstream_devices_error_reporting(pdev, false);
672 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
673 /* Disable Root's interrupt in response to error messages */
674 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_COMMAND, 0);
676 /* Clear Root's error status reg */
677 pci_read_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, ®32);
678 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, reg32);
682 * get_e_source - retrieve an error source
683 * @rpc: pointer to the root port which holds an error
685 * Invoked by DPC handler to consume an error.
687 static struct aer_err_source *get_e_source(struct aer_rpc *rpc)
689 struct aer_err_source *e_source;
692 /* Lock access to Root error producer/consumer index */
693 spin_lock_irqsave(&rpc->e_lock, flags);
694 if (rpc->prod_idx == rpc->cons_idx) {
695 spin_unlock_irqrestore(&rpc->e_lock, flags);
698 e_source = &rpc->e_sources[rpc->cons_idx];
700 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
702 spin_unlock_irqrestore(&rpc->e_lock, flags);
707 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
711 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
713 /* The device might not support AER */
717 if (info->severity == AER_CORRECTABLE) {
718 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
720 if (!(info->status & ERR_CORRECTABLE_ERROR_MASK))
721 return AER_UNSUCCESS;
722 } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
723 info->severity == AER_NONFATAL) {
725 /* Link is still healthy for IO reads */
726 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
728 if (!(info->status & ERR_UNCORRECTABLE_ERROR_MASK))
729 return AER_UNSUCCESS;
731 if (info->status & AER_LOG_TLP_MASKS) {
732 info->flags |= AER_TLP_HEADER_VALID_FLAG;
733 pci_read_config_dword(dev,
734 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
735 pci_read_config_dword(dev,
736 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
737 pci_read_config_dword(dev,
738 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
739 pci_read_config_dword(dev,
740 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
747 static inline void aer_process_err_devices(struct pcie_device *p_device,
748 struct aer_err_info *e_info)
752 if (!e_info->dev[0]) {
753 dev_printk(KERN_DEBUG, &p_device->port->dev,
754 "can't find device of ID%04x\n",
758 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
759 if (get_device_error_info(e_info->dev[i], e_info) ==
761 aer_print_error(e_info->dev[i], e_info);
762 handle_error_source(p_device,
770 * aer_isr_one_error - consume an error detected by root port
771 * @p_device: pointer to error root port service device
772 * @e_src: pointer to an error source
774 static void aer_isr_one_error(struct pcie_device *p_device,
775 struct aer_err_source *e_src)
777 struct aer_err_info *e_info;
780 /* struct aer_err_info might be big, so we allocate it with slab */
781 e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL);
782 if (e_info == NULL) {
783 dev_printk(KERN_DEBUG, &p_device->port->dev,
784 "Can't allocate mem when processing AER errors\n");
789 * There is a possibility that both correctable error and
790 * uncorrectable error being logged. Report correctable error first.
792 for (i = 1; i & ROOT_ERR_STATUS_MASKS ; i <<= 2) {
795 if (!(e_src->status & i))
798 memset(e_info, 0, sizeof(struct aer_err_info));
800 /* Init comprehensive error information */
801 if (i & PCI_ERR_ROOT_COR_RCV) {
802 e_info->id = ERR_COR_ID(e_src->id);
803 e_info->severity = AER_CORRECTABLE;
805 e_info->id = ERR_UNCOR_ID(e_src->id);
806 e_info->severity = ((e_src->status >> 6) & 1);
809 (PCI_ERR_ROOT_MULTI_COR_RCV |
810 PCI_ERR_ROOT_MULTI_UNCOR_RCV))
811 e_info->flags |= AER_MULTI_ERROR_VALID_FLAG;
813 find_source_device(p_device->port, e_info);
814 aer_process_err_devices(p_device, e_info);
821 * aer_isr - consume errors detected by root port
822 * @work: definition of this work item
824 * Invoked, as DPC, when root port records new detected error
826 void aer_isr(struct work_struct *work)
828 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
829 struct pcie_device *p_device = rpc->rpd;
830 struct aer_err_source *e_src;
832 mutex_lock(&rpc->rpc_mutex);
833 e_src = get_e_source(rpc);
835 aer_isr_one_error(p_device, e_src);
836 e_src = get_e_source(rpc);
838 mutex_unlock(&rpc->rpc_mutex);
840 wake_up(&rpc->wait_release);
844 * aer_delete_rootport - disable root port aer and delete service data
845 * @rpc: pointer to a root port device being deleted
847 * Invoked when AER service unloaded on a specific Root Port
849 void aer_delete_rootport(struct aer_rpc *rpc)
851 /* Disable root port AER itself */
852 disable_root_aer(rpc);
858 * aer_init - provide AER initialization
859 * @dev: pointer to AER pcie device
861 * Invoked when AER service driver is loaded.
863 int aer_init(struct pcie_device *dev)
865 if (aer_osc_setup(dev) && !forceload)