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 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
36 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
40 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
44 pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, ®16);
47 PCI_EXP_DEVCTL_NFERE |
50 pci_write_config_word(dev, pos+PCI_EXP_DEVCTL,
55 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
60 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
64 pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, ®16);
65 reg16 = reg16 & ~(PCI_EXP_DEVCTL_CERE |
66 PCI_EXP_DEVCTL_NFERE |
69 pci_write_config_word(dev, pos+PCI_EXP_DEVCTL,
74 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
79 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
83 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
84 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
85 if (dev->error_state == pci_channel_io_normal)
86 status &= ~mask; /* Clear corresponding nonfatal bits */
88 status &= mask; /* Clear corresponding fatal bits */
89 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
95 int pci_cleanup_aer_correct_error_status(struct pci_dev *dev)
100 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
104 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
105 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
112 static int set_device_error_reporting(struct pci_dev *dev, void *data)
114 bool enable = *((bool *)data);
116 if (dev->pcie_type == PCIE_RC_PORT ||
117 dev->pcie_type == PCIE_SW_UPSTREAM_PORT ||
118 dev->pcie_type == PCIE_SW_DOWNSTREAM_PORT) {
120 pci_enable_pcie_error_reporting(dev);
122 pci_disable_pcie_error_reporting(dev);
126 pcie_set_ecrc_checking(dev);
132 * set_downstream_devices_error_reporting - enable/disable the error reporting bits on the root port and its downstream ports.
133 * @dev: pointer to root port's pci_dev data structure
134 * @enable: true = enable error reporting, false = disable error reporting.
136 static void set_downstream_devices_error_reporting(struct pci_dev *dev,
139 set_device_error_reporting(dev, &enable);
141 if (!dev->subordinate)
143 pci_walk_bus(dev->subordinate, set_device_error_reporting, &enable);
146 static int find_device_iter(struct device *device, void *data)
149 u16 id = *(unsigned long *)data;
150 u8 secondary, subordinate, d_bus = id >> 8;
152 if (device->bus == &pci_bus_type) {
153 dev = to_pci_dev(device);
154 if (id == ((dev->bus->number << 8) | dev->devfn)) {
158 *(unsigned long*)data = (unsigned long)device;
163 * If device is P2P, check if it is an upstream?
165 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
166 pci_read_config_byte(dev, PCI_SECONDARY_BUS,
168 pci_read_config_byte(dev, PCI_SUBORDINATE_BUS,
170 if (d_bus >= secondary && d_bus <= subordinate) {
171 *(unsigned long*)data = (unsigned long)device;
181 * find_source_device - search through device hierarchy for source device
182 * @parent: pointer to Root Port pci_dev data structure
183 * @id: device ID of agent who sends an error message to this Root Port
185 * Invoked when error is detected at the Root Port.
187 static struct device* find_source_device(struct pci_dev *parent, u16 id)
189 struct pci_dev *dev = parent;
190 struct device *device;
191 unsigned long device_addr;
194 /* Is Root Port an agent that sends error message? */
195 if (id == ((dev->bus->number << 8) | dev->devfn))
200 if ((status = device_for_each_child(&dev->dev,
201 &device_addr, find_device_iter))) {
202 device = (struct device*)device_addr;
203 dev = to_pci_dev(device);
204 if (id == ((dev->bus->number << 8) | dev->devfn))
212 static int report_error_detected(struct pci_dev *dev, void *data)
214 pci_ers_result_t vote;
215 struct pci_error_handlers *err_handler;
216 struct aer_broadcast_data *result_data;
217 result_data = (struct aer_broadcast_data *) data;
219 dev->error_state = result_data->state;
222 !dev->driver->err_handler ||
223 !dev->driver->err_handler->error_detected) {
224 if (result_data->state == pci_channel_io_frozen &&
225 !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
227 * In case of fatal recovery, if one of down-
228 * stream device has no driver. We might be
229 * unable to recover because a later insmod
230 * of a driver for this device is unaware of
233 dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
235 "no AER-aware driver" : "no driver");
240 err_handler = dev->driver->err_handler;
241 vote = err_handler->error_detected(dev, result_data->state);
242 result_data->result = merge_result(result_data->result, vote);
246 static int report_mmio_enabled(struct pci_dev *dev, void *data)
248 pci_ers_result_t vote;
249 struct pci_error_handlers *err_handler;
250 struct aer_broadcast_data *result_data;
251 result_data = (struct aer_broadcast_data *) data;
254 !dev->driver->err_handler ||
255 !dev->driver->err_handler->mmio_enabled)
258 err_handler = dev->driver->err_handler;
259 vote = err_handler->mmio_enabled(dev);
260 result_data->result = merge_result(result_data->result, vote);
264 static int report_slot_reset(struct pci_dev *dev, void *data)
266 pci_ers_result_t vote;
267 struct pci_error_handlers *err_handler;
268 struct aer_broadcast_data *result_data;
269 result_data = (struct aer_broadcast_data *) data;
272 !dev->driver->err_handler ||
273 !dev->driver->err_handler->slot_reset)
276 err_handler = dev->driver->err_handler;
277 vote = err_handler->slot_reset(dev);
278 result_data->result = merge_result(result_data->result, vote);
282 static int report_resume(struct pci_dev *dev, void *data)
284 struct pci_error_handlers *err_handler;
286 dev->error_state = pci_channel_io_normal;
289 !dev->driver->err_handler ||
290 !dev->driver->err_handler->resume)
293 err_handler = dev->driver->err_handler;
294 err_handler->resume(dev);
299 * broadcast_error_message - handle message broadcast to downstream drivers
300 * @dev: pointer to from where in a hierarchy message is broadcasted down
301 * @state: error state
302 * @error_mesg: message to print
303 * @cb: callback to be broadcasted
305 * Invoked during error recovery process. Once being invoked, the content
306 * of error severity will be broadcasted to all downstream drivers in a
307 * hierarchy in question.
309 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
310 enum pci_channel_state state,
312 int (*cb)(struct pci_dev *, void *))
314 struct aer_broadcast_data result_data;
316 dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
317 result_data.state = state;
318 if (cb == report_error_detected)
319 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
321 result_data.result = PCI_ERS_RESULT_RECOVERED;
323 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
325 * If the error is reported by a bridge, we think this error
326 * is related to the downstream link of the bridge, so we
327 * do error recovery on all subordinates of the bridge instead
328 * of the bridge and clear the error status of the bridge.
330 if (cb == report_error_detected)
331 dev->error_state = state;
332 pci_walk_bus(dev->subordinate, cb, &result_data);
333 if (cb == report_resume) {
334 pci_cleanup_aer_uncorrect_error_status(dev);
335 dev->error_state = pci_channel_io_normal;
340 * If the error is reported by an end point, we think this
341 * error is related to the upstream link of the end point.
343 pci_walk_bus(dev->bus, cb, &result_data);
346 return result_data.result;
349 struct find_aer_service_data {
350 struct pcie_port_service_driver *aer_driver;
354 static int find_aer_service_iter(struct device *device, void *data)
356 struct device_driver *driver;
357 struct pcie_port_service_driver *service_driver;
358 struct find_aer_service_data *result;
360 result = (struct find_aer_service_data *) data;
362 if (device->bus == &pcie_port_bus_type) {
363 struct pcie_port_data *port_data;
365 port_data = pci_get_drvdata(to_pcie_device(device)->port);
366 if (port_data->port_type == PCIE_SW_DOWNSTREAM_PORT)
367 result->is_downstream = 1;
369 driver = device->driver;
371 service_driver = to_service_driver(driver);
372 if (service_driver->service == PCIE_PORT_SERVICE_AER) {
373 result->aer_driver = service_driver;
382 static void find_aer_service(struct pci_dev *dev,
383 struct find_aer_service_data *data)
386 retval = device_for_each_child(&dev->dev, data, find_aer_service_iter);
389 static pci_ers_result_t reset_link(struct pcie_device *aerdev,
392 struct pci_dev *udev;
393 pci_ers_result_t status;
394 struct find_aer_service_data data;
396 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)
399 udev= dev->bus->self;
401 data.is_downstream = 0;
402 data.aer_driver = NULL;
403 find_aer_service(udev, &data);
406 * Use the aer driver of the error agent firstly.
407 * If it hasn't the aer driver, use the root port's
409 if (!data.aer_driver || !data.aer_driver->reset_link) {
410 if (data.is_downstream &&
411 aerdev->device.driver &&
412 to_service_driver(aerdev->device.driver)->reset_link) {
414 to_service_driver(aerdev->device.driver);
416 dev_printk(KERN_DEBUG, &dev->dev, "no link-reset "
418 return PCI_ERS_RESULT_DISCONNECT;
422 status = data.aer_driver->reset_link(udev);
423 if (status != PCI_ERS_RESULT_RECOVERED) {
424 dev_printk(KERN_DEBUG, &dev->dev, "link reset at upstream "
425 "device %s failed\n", pci_name(udev));
426 return PCI_ERS_RESULT_DISCONNECT;
433 * do_recovery - handle nonfatal/fatal error recovery process
434 * @aerdev: pointer to a pcie_device data structure of root port
435 * @dev: pointer to a pci_dev data structure of agent detecting an error
436 * @severity: error severity type
438 * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
439 * error detected message to all downstream drivers within a hierarchy in
440 * question and return the returned code.
442 static pci_ers_result_t do_recovery(struct pcie_device *aerdev,
446 pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
447 enum pci_channel_state state;
449 if (severity == AER_FATAL)
450 state = pci_channel_io_frozen;
452 state = pci_channel_io_normal;
454 status = broadcast_error_message(dev,
457 report_error_detected);
459 if (severity == AER_FATAL) {
460 result = reset_link(aerdev, dev);
461 if (result != PCI_ERS_RESULT_RECOVERED) {
462 /* TODO: Should panic here? */
467 if (status == PCI_ERS_RESULT_CAN_RECOVER)
468 status = broadcast_error_message(dev,
471 report_mmio_enabled);
473 if (status == PCI_ERS_RESULT_NEED_RESET) {
475 * TODO: Should call platform-specific
476 * functions to reset slot before calling
477 * drivers' slot_reset callbacks?
479 status = broadcast_error_message(dev,
485 if (status == PCI_ERS_RESULT_RECOVERED)
486 broadcast_error_message(dev,
495 * handle_error_source - handle logging error into an event log
496 * @aerdev: pointer to pcie_device data structure of the root port
497 * @dev: pointer to pci_dev data structure of error source device
498 * @info: comprehensive error information
500 * Invoked when an error being detected by Root Port.
502 static void handle_error_source(struct pcie_device * aerdev,
504 struct aer_err_info info)
506 pci_ers_result_t status = 0;
509 if (info.severity == AER_CORRECTABLE) {
511 * Correctable error does not need software intevention.
512 * No need to go through error recovery process.
514 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
516 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
519 status = do_recovery(aerdev, dev, info.severity);
520 if (status == PCI_ERS_RESULT_RECOVERED) {
521 dev_printk(KERN_DEBUG, &dev->dev, "AER driver "
522 "successfully recovered\n");
524 /* TODO: Should kernel panic here? */
525 dev_printk(KERN_DEBUG, &dev->dev, "AER driver didn't "
532 * aer_enable_rootport - enable Root Port's interrupts when receiving messages
533 * @rpc: pointer to a Root Port data structure
535 * Invoked when PCIE bus loads AER service driver.
537 void aer_enable_rootport(struct aer_rpc *rpc)
539 struct pci_dev *pdev = rpc->rpd->port;
544 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
545 /* Clear PCIE Capability's Device Status */
546 pci_read_config_word(pdev, pos+PCI_EXP_DEVSTA, ®16);
547 pci_write_config_word(pdev, pos+PCI_EXP_DEVSTA, reg16);
549 /* Disable system error generation in response to error messages */
550 pci_read_config_word(pdev, pos + PCI_EXP_RTCTL, ®16);
551 reg16 &= ~(SYSTEM_ERROR_INTR_ON_MESG_MASK);
552 pci_write_config_word(pdev, pos + PCI_EXP_RTCTL, reg16);
554 aer_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
555 /* Clear error status */
556 pci_read_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, ®32);
557 pci_write_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, reg32);
558 pci_read_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, ®32);
559 pci_write_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, reg32);
560 pci_read_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, ®32);
561 pci_write_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, reg32);
564 * Enable error reporting for the root port device and downstream port
567 set_downstream_devices_error_reporting(pdev, true);
569 /* Enable Root Port's interrupt in response to error messages */
570 pci_write_config_dword(pdev,
571 aer_pos + PCI_ERR_ROOT_COMMAND,
572 ROOT_PORT_INTR_ON_MESG_MASK);
576 * disable_root_aer - disable Root Port's interrupts when receiving messages
577 * @rpc: pointer to a Root Port data structure
579 * Invoked when PCIE bus unloads AER service driver.
581 static void disable_root_aer(struct aer_rpc *rpc)
583 struct pci_dev *pdev = rpc->rpd->port;
588 * Disable error reporting for the root port device and downstream port
591 set_downstream_devices_error_reporting(pdev, false);
593 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
594 /* Disable Root's interrupt in response to error messages */
595 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_COMMAND, 0);
597 /* Clear Root's error status reg */
598 pci_read_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, ®32);
599 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, reg32);
603 * get_e_source - retrieve an error source
604 * @rpc: pointer to the root port which holds an error
606 * Invoked by DPC handler to consume an error.
608 static struct aer_err_source* get_e_source(struct aer_rpc *rpc)
610 struct aer_err_source *e_source;
613 /* Lock access to Root error producer/consumer index */
614 spin_lock_irqsave(&rpc->e_lock, flags);
615 if (rpc->prod_idx == rpc->cons_idx) {
616 spin_unlock_irqrestore(&rpc->e_lock, flags);
619 e_source = &rpc->e_sources[rpc->cons_idx];
621 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
623 spin_unlock_irqrestore(&rpc->e_lock, flags);
628 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
632 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
634 /* The device might not support AER */
638 if (info->severity == AER_CORRECTABLE) {
639 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
641 if (!(info->status & ERR_CORRECTABLE_ERROR_MASK))
642 return AER_UNSUCCESS;
643 } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
644 info->severity == AER_NONFATAL) {
646 /* Link is still healthy for IO reads */
647 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
649 if (!(info->status & ERR_UNCORRECTABLE_ERROR_MASK))
650 return AER_UNSUCCESS;
652 if (info->status & AER_LOG_TLP_MASKS) {
653 info->flags |= AER_TLP_HEADER_VALID_FLAG;
654 pci_read_config_dword(dev,
655 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
656 pci_read_config_dword(dev,
657 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
658 pci_read_config_dword(dev,
659 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
660 pci_read_config_dword(dev,
661 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
669 * aer_isr_one_error - consume an error detected by root port
670 * @p_device: pointer to error root port service device
671 * @e_src: pointer to an error source
673 static void aer_isr_one_error(struct pcie_device *p_device,
674 struct aer_err_source *e_src)
676 struct device *s_device;
677 struct aer_err_info e_info = {0, 0, 0,};
682 * There is a possibility that both correctable error and
683 * uncorrectable error being logged. Report correctable error first.
685 for (i = 1; i & ROOT_ERR_STATUS_MASKS ; i <<= 2) {
688 if (!(e_src->status & i))
691 /* Init comprehensive error information */
692 if (i & PCI_ERR_ROOT_COR_RCV) {
693 id = ERR_COR_ID(e_src->id);
694 e_info.severity = AER_CORRECTABLE;
696 id = ERR_UNCOR_ID(e_src->id);
697 e_info.severity = ((e_src->status >> 6) & 1);
700 (PCI_ERR_ROOT_MULTI_COR_RCV |
701 PCI_ERR_ROOT_MULTI_UNCOR_RCV))
702 e_info.flags |= AER_MULTI_ERROR_VALID_FLAG;
703 if (!(s_device = find_source_device(p_device->port, id))) {
704 printk(KERN_DEBUG "%s->can't find device of ID%04x\n",
708 if (get_device_error_info(to_pci_dev(s_device), &e_info) ==
710 aer_print_error(to_pci_dev(s_device), &e_info);
711 handle_error_source(p_device,
712 to_pci_dev(s_device),
719 * aer_isr - consume errors detected by root port
720 * @work: definition of this work item
722 * Invoked, as DPC, when root port records new detected error
724 void aer_isr(struct work_struct *work)
726 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
727 struct pcie_device *p_device = rpc->rpd;
728 struct aer_err_source *e_src;
730 mutex_lock(&rpc->rpc_mutex);
731 e_src = get_e_source(rpc);
733 aer_isr_one_error(p_device, e_src);
734 e_src = get_e_source(rpc);
736 mutex_unlock(&rpc->rpc_mutex);
738 wake_up(&rpc->wait_release);
742 * aer_delete_rootport - disable root port aer and delete service data
743 * @rpc: pointer to a root port device being deleted
745 * Invoked when AER service unloaded on a specific Root Port
747 void aer_delete_rootport(struct aer_rpc *rpc)
749 /* Disable root port AER itself */
750 disable_root_aer(rpc);
756 * aer_init - provide AER initialization
757 * @dev: pointer to AER pcie device
759 * Invoked when AER service driver is loaded.
761 int aer_init(struct pcie_device *dev)
763 if (aer_osc_setup(dev) && !forceload)
769 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
770 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
771 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);