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);
121 int pci_cleanup_aer_correct_error_status(struct pci_dev *dev)
126 pos = pci_find_aer_capability(dev);
130 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
131 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
137 static int find_device_iter(struct device *device, void *data)
140 u16 id = *(unsigned long *)data;
141 u8 secondary, subordinate, d_bus = id >> 8;
143 if (device->bus == &pci_bus_type) {
144 dev = to_pci_dev(device);
145 if (id == ((dev->bus->number << 8) | dev->devfn)) {
149 *(unsigned long*)data = (unsigned long)device;
154 * If device is P2P, check if it is an upstream?
156 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
157 pci_read_config_byte(dev, PCI_SECONDARY_BUS,
159 pci_read_config_byte(dev, PCI_SUBORDINATE_BUS,
161 if (d_bus >= secondary && d_bus <= subordinate) {
162 *(unsigned long*)data = (unsigned long)device;
172 * find_source_device - search through device hierarchy for source device
173 * @parent: pointer to Root Port pci_dev data structure
174 * @id: device ID of agent who sends an error message to this Root Port
176 * Invoked when error is detected at the Root Port.
178 static struct device* find_source_device(struct pci_dev *parent, u16 id)
180 struct pci_dev *dev = parent;
181 struct device *device;
182 unsigned long device_addr;
185 /* Is Root Port an agent that sends error message? */
186 if (id == ((dev->bus->number << 8) | dev->devfn))
191 if ((status = device_for_each_child(&dev->dev,
192 &device_addr, find_device_iter))) {
193 device = (struct device*)device_addr;
194 dev = to_pci_dev(device);
195 if (id == ((dev->bus->number << 8) | dev->devfn))
203 static void report_error_detected(struct pci_dev *dev, void *data)
205 pci_ers_result_t vote;
206 struct pci_error_handlers *err_handler;
207 struct aer_broadcast_data *result_data;
208 result_data = (struct aer_broadcast_data *) data;
210 dev->error_state = result_data->state;
213 !dev->driver->err_handler ||
214 !dev->driver->err_handler->error_detected) {
215 if (result_data->state == pci_channel_io_frozen &&
216 !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
218 * In case of fatal recovery, if one of down-
219 * stream device has no driver. We might be
220 * unable to recover because a later insmod
221 * of a driver for this device is unaware of
224 dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
226 "no AER-aware driver" : "no driver");
231 err_handler = dev->driver->err_handler;
232 vote = err_handler->error_detected(dev, result_data->state);
233 result_data->result = merge_result(result_data->result, vote);
237 static void report_mmio_enabled(struct pci_dev *dev, void *data)
239 pci_ers_result_t vote;
240 struct pci_error_handlers *err_handler;
241 struct aer_broadcast_data *result_data;
242 result_data = (struct aer_broadcast_data *) data;
245 !dev->driver->err_handler ||
246 !dev->driver->err_handler->mmio_enabled)
249 err_handler = dev->driver->err_handler;
250 vote = err_handler->mmio_enabled(dev);
251 result_data->result = merge_result(result_data->result, vote);
255 static void report_slot_reset(struct pci_dev *dev, void *data)
257 pci_ers_result_t vote;
258 struct pci_error_handlers *err_handler;
259 struct aer_broadcast_data *result_data;
260 result_data = (struct aer_broadcast_data *) data;
263 !dev->driver->err_handler ||
264 !dev->driver->err_handler->slot_reset)
267 err_handler = dev->driver->err_handler;
268 vote = err_handler->slot_reset(dev);
269 result_data->result = merge_result(result_data->result, vote);
273 static void report_resume(struct pci_dev *dev, void *data)
275 struct pci_error_handlers *err_handler;
277 dev->error_state = pci_channel_io_normal;
280 !dev->driver->err_handler ||
281 !dev->driver->err_handler->slot_reset)
284 err_handler = dev->driver->err_handler;
285 err_handler->resume(dev);
290 * broadcast_error_message - handle message broadcast to downstream drivers
291 * @dev: pointer to from where in a hierarchy message is broadcasted down
292 * @state: error state
293 * @error_mesg: message to print
294 * @cb: callback to be broadcasted
296 * Invoked during error recovery process. Once being invoked, the content
297 * of error severity will be broadcasted to all downstream drivers in a
298 * hierarchy in question.
300 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
301 enum pci_channel_state state,
303 void (*cb)(struct pci_dev *, void *))
305 struct aer_broadcast_data result_data;
307 dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
308 result_data.state = state;
309 if (cb == report_error_detected)
310 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
312 result_data.result = PCI_ERS_RESULT_RECOVERED;
314 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
316 * If the error is reported by a bridge, we think this error
317 * is related to the downstream link of the bridge, so we
318 * do error recovery on all subordinates of the bridge instead
319 * of the bridge and clear the error status of the bridge.
321 if (cb == report_error_detected)
322 dev->error_state = state;
323 pci_walk_bus(dev->subordinate, cb, &result_data);
324 if (cb == report_resume) {
325 pci_cleanup_aer_uncorrect_error_status(dev);
326 dev->error_state = pci_channel_io_normal;
331 * If the error is reported by an end point, we think this
332 * error is related to the upstream link of the end point.
334 pci_walk_bus(dev->bus, cb, &result_data);
337 return result_data.result;
340 struct find_aer_service_data {
341 struct pcie_port_service_driver *aer_driver;
345 static int find_aer_service_iter(struct device *device, void *data)
347 struct device_driver *driver;
348 struct pcie_port_service_driver *service_driver;
349 struct pcie_device *pcie_dev;
350 struct find_aer_service_data *result;
352 result = (struct find_aer_service_data *) data;
354 if (device->bus == &pcie_port_bus_type) {
355 pcie_dev = to_pcie_device(device);
356 if (pcie_dev->id.port_type == PCIE_SW_DOWNSTREAM_PORT)
357 result->is_downstream = 1;
359 driver = device->driver;
361 service_driver = to_service_driver(driver);
362 if (service_driver->id_table->service_type ==
363 PCIE_PORT_SERVICE_AER) {
364 result->aer_driver = service_driver;
373 static void find_aer_service(struct pci_dev *dev,
374 struct find_aer_service_data *data)
377 retval = device_for_each_child(&dev->dev, data, find_aer_service_iter);
380 static pci_ers_result_t reset_link(struct pcie_device *aerdev,
383 struct pci_dev *udev;
384 pci_ers_result_t status;
385 struct find_aer_service_data data;
387 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)
390 udev= dev->bus->self;
392 data.is_downstream = 0;
393 data.aer_driver = NULL;
394 find_aer_service(udev, &data);
397 * Use the aer driver of the error agent firstly.
398 * If it hasn't the aer driver, use the root port's
400 if (!data.aer_driver || !data.aer_driver->reset_link) {
401 if (data.is_downstream &&
402 aerdev->device.driver &&
403 to_service_driver(aerdev->device.driver)->reset_link) {
405 to_service_driver(aerdev->device.driver);
407 dev_printk(KERN_DEBUG, &dev->dev, "no link-reset "
409 return PCI_ERS_RESULT_DISCONNECT;
413 status = data.aer_driver->reset_link(udev);
414 if (status != PCI_ERS_RESULT_RECOVERED) {
415 dev_printk(KERN_DEBUG, &dev->dev, "link reset at upstream "
416 "device %s failed\n", pci_name(udev));
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 dev_printk(KERN_DEBUG, &dev->dev, "AER driver "
513 "successfully recovered\n");
515 /* TODO: Should kernel panic here? */
516 dev_printk(KERN_DEBUG, &dev->dev, "AER driver didn't "
523 * aer_enable_rootport - enable Root Port's interrupts when receiving messages
524 * @rpc: pointer to a Root Port data structure
526 * Invoked when PCIE bus loads AER service driver.
528 void aer_enable_rootport(struct aer_rpc *rpc)
530 struct pci_dev *pdev = rpc->rpd->port;
535 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
536 /* Clear PCIE Capability's Device Status */
537 pci_read_config_word(pdev, pos+PCI_EXP_DEVSTA, ®16);
538 pci_write_config_word(pdev, pos+PCI_EXP_DEVSTA, reg16);
540 /* Disable system error generation in response to error messages */
541 pci_read_config_word(pdev, pos + PCI_EXP_RTCTL, ®16);
542 reg16 &= ~(SYSTEM_ERROR_INTR_ON_MESG_MASK);
543 pci_write_config_word(pdev, pos + PCI_EXP_RTCTL, reg16);
545 aer_pos = pci_find_aer_capability(pdev);
546 /* Clear error status */
547 pci_read_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, ®32);
548 pci_write_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, reg32);
549 pci_read_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, ®32);
550 pci_write_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, reg32);
551 pci_read_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, ®32);
552 pci_write_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, reg32);
554 /* Enable Root Port device reporting error itself */
555 pci_read_config_word(pdev, pos+PCI_EXP_DEVCTL, ®16);
557 PCI_EXP_DEVCTL_CERE |
558 PCI_EXP_DEVCTL_NFERE |
559 PCI_EXP_DEVCTL_FERE |
561 pci_write_config_word(pdev, pos+PCI_EXP_DEVCTL,
564 /* Enable Root Port's interrupt in response to error messages */
565 pci_write_config_dword(pdev,
566 aer_pos + PCI_ERR_ROOT_COMMAND,
567 ROOT_PORT_INTR_ON_MESG_MASK);
571 * disable_root_aer - disable Root Port's interrupts when receiving messages
572 * @rpc: pointer to a Root Port data structure
574 * Invoked when PCIE bus unloads AER service driver.
576 static void disable_root_aer(struct aer_rpc *rpc)
578 struct pci_dev *pdev = rpc->rpd->port;
582 pos = pci_find_aer_capability(pdev);
583 /* Disable Root's interrupt in response to error messages */
584 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_COMMAND, 0);
586 /* Clear Root's error status reg */
587 pci_read_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, ®32);
588 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, reg32);
592 * get_e_source - retrieve an error source
593 * @rpc: pointer to the root port which holds an error
595 * Invoked by DPC handler to consume an error.
597 static struct aer_err_source* get_e_source(struct aer_rpc *rpc)
599 struct aer_err_source *e_source;
602 /* Lock access to Root error producer/consumer index */
603 spin_lock_irqsave(&rpc->e_lock, flags);
604 if (rpc->prod_idx == rpc->cons_idx) {
605 spin_unlock_irqrestore(&rpc->e_lock, flags);
608 e_source = &rpc->e_sources[rpc->cons_idx];
610 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
612 spin_unlock_irqrestore(&rpc->e_lock, flags);
617 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
621 pos = pci_find_aer_capability(dev);
623 /* The device might not support AER */
627 if (info->severity == AER_CORRECTABLE) {
628 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
630 if (!(info->status & ERR_CORRECTABLE_ERROR_MASK))
631 return AER_UNSUCCESS;
632 } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
633 info->severity == AER_NONFATAL) {
635 /* Link is still healthy for IO reads */
636 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
638 if (!(info->status & ERR_UNCORRECTABLE_ERROR_MASK))
639 return AER_UNSUCCESS;
641 if (info->status & AER_LOG_TLP_MASKS) {
642 info->flags |= AER_TLP_HEADER_VALID_FLAG;
643 pci_read_config_dword(dev,
644 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
645 pci_read_config_dword(dev,
646 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
647 pci_read_config_dword(dev,
648 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
649 pci_read_config_dword(dev,
650 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
658 * aer_isr_one_error - consume an error detected by root port
659 * @p_device: pointer to error root port service device
660 * @e_src: pointer to an error source
662 static void aer_isr_one_error(struct pcie_device *p_device,
663 struct aer_err_source *e_src)
665 struct device *s_device;
666 struct aer_err_info e_info = {0, 0, 0,};
671 * There is a possibility that both correctable error and
672 * uncorrectable error being logged. Report correctable error first.
674 for (i = 1; i & ROOT_ERR_STATUS_MASKS ; i <<= 2) {
677 if (!(e_src->status & i))
680 /* Init comprehensive error information */
681 if (i & PCI_ERR_ROOT_COR_RCV) {
682 id = ERR_COR_ID(e_src->id);
683 e_info.severity = AER_CORRECTABLE;
685 id = ERR_UNCOR_ID(e_src->id);
686 e_info.severity = ((e_src->status >> 6) & 1);
689 (PCI_ERR_ROOT_MULTI_COR_RCV |
690 PCI_ERR_ROOT_MULTI_UNCOR_RCV))
691 e_info.flags |= AER_MULTI_ERROR_VALID_FLAG;
692 if (!(s_device = find_source_device(p_device->port, id))) {
693 printk(KERN_DEBUG "%s->can't find device of ID%04x\n",
697 if (get_device_error_info(to_pci_dev(s_device), &e_info) ==
699 aer_print_error(to_pci_dev(s_device), &e_info);
700 handle_error_source(p_device,
701 to_pci_dev(s_device),
708 * aer_isr - consume errors detected by root port
709 * @work: definition of this work item
711 * Invoked, as DPC, when root port records new detected error
713 void aer_isr(struct work_struct *work)
715 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
716 struct pcie_device *p_device = rpc->rpd;
717 struct aer_err_source *e_src;
719 mutex_lock(&rpc->rpc_mutex);
720 e_src = get_e_source(rpc);
722 aer_isr_one_error(p_device, e_src);
723 e_src = get_e_source(rpc);
725 mutex_unlock(&rpc->rpc_mutex);
727 wake_up(&rpc->wait_release);
731 * aer_delete_rootport - disable root port aer and delete service data
732 * @rpc: pointer to a root port device being deleted
734 * Invoked when AER service unloaded on a specific Root Port
736 void aer_delete_rootport(struct aer_rpc *rpc)
738 /* Disable root port AER itself */
739 disable_root_aer(rpc);
745 * aer_init - provide AER initialization
746 * @dev: pointer to AER pcie device
748 * Invoked when AER service driver is loaded.
750 int aer_init(struct pcie_device *dev)
752 if (aer_osc_setup(dev) && !forceload)
758 EXPORT_SYMBOL_GPL(pci_find_aer_capability);
759 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
760 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
761 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);