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 PCIe AER. When a PCIe
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>
26 #include <linux/slab.h>
27 #include <linux/kfifo.h>
30 #define PCI_EXP_AER_FLAGS (PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \
31 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE)
33 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
35 if (pcie_aer_get_firmware_first(dev))
41 return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
43 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
45 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
47 if (pcie_aer_get_firmware_first(dev))
50 return pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
53 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
55 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
64 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
66 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
70 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
72 int pci_cleanup_aer_error_status_regs(struct pci_dev *dev)
78 if (!pci_is_pcie(dev))
85 port_type = pci_pcie_type(dev);
86 if (port_type == PCI_EXP_TYPE_ROOT_PORT) {
87 pci_read_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, &status);
88 pci_write_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, status);
91 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
92 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
94 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
95 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
100 int pci_aer_init(struct pci_dev *dev)
102 dev->aer_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
103 return pci_cleanup_aer_error_status_regs(dev);
107 * add_error_device - list device to be handled
108 * @e_info: pointer to error info
109 * @dev: pointer to pci_dev to be added
111 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
113 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
114 e_info->dev[e_info->error_dev_num] = dev;
115 e_info->error_dev_num++;
122 * is_error_source - check whether the device is source of reported error
123 * @dev: pointer to pci_dev to be checked
124 * @e_info: pointer to reported error info
126 static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info)
133 * When bus id is equal to 0, it might be a bad id
134 * reported by root port.
136 if ((PCI_BUS_NUM(e_info->id) != 0) &&
137 !(dev->bus->bus_flags & PCI_BUS_FLAGS_NO_AERSID)) {
138 /* Device ID match? */
139 if (e_info->id == ((dev->bus->number << 8) | dev->devfn))
142 /* Continue id comparing if there is no multiple error */
143 if (!e_info->multi_error_valid)
149 * 1) bus id is equal to 0. Some ports might lose the bus
150 * id of error source id;
151 * 2) bus flag PCI_BUS_FLAGS_NO_AERSID is set
152 * 3) There are multiple errors and prior ID comparing fails;
153 * We check AER status registers to find possible reporter.
155 if (atomic_read(&dev->enable_cnt) == 0)
158 /* Check if AER is enabled */
159 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, ®16);
160 if (!(reg16 & PCI_EXP_AER_FLAGS))
167 /* Check if error is recorded */
168 if (e_info->severity == AER_CORRECTABLE) {
169 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
170 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, &mask);
172 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
173 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
181 static int find_device_iter(struct pci_dev *dev, void *data)
183 struct aer_err_info *e_info = (struct aer_err_info *)data;
185 if (is_error_source(dev, e_info)) {
186 /* List this device */
187 if (add_error_device(e_info, dev)) {
188 /* We cannot handle more... Stop iteration */
189 /* TODO: Should print error message here? */
193 /* If there is only a single error, stop iteration */
194 if (!e_info->multi_error_valid)
201 * find_source_device - search through device hierarchy for source device
202 * @parent: pointer to Root Port pci_dev data structure
203 * @e_info: including detailed error information such like id
205 * Return true if found.
207 * Invoked by DPC when error is detected at the Root Port.
208 * Caller of this function must set id, severity, and multi_error_valid of
209 * struct aer_err_info pointed by @e_info properly. This function must fill
210 * e_info->error_dev_num and e_info->dev[], based on the given information.
212 static bool find_source_device(struct pci_dev *parent,
213 struct aer_err_info *e_info)
215 struct pci_dev *dev = parent;
218 /* Must reset in this function */
219 e_info->error_dev_num = 0;
221 /* Is Root Port an agent that sends error message? */
222 result = find_device_iter(dev, e_info);
226 pci_walk_bus(parent->subordinate, find_device_iter, e_info);
228 if (!e_info->error_dev_num) {
229 dev_printk(KERN_DEBUG, &parent->dev,
230 "can't find device of ID%04x\n",
237 static int report_error_detected(struct pci_dev *dev, void *data)
239 pci_ers_result_t vote;
240 const struct pci_error_handlers *err_handler;
241 struct aer_broadcast_data *result_data;
242 result_data = (struct aer_broadcast_data *) data;
244 device_lock(&dev->dev);
245 dev->error_state = result_data->state;
248 !dev->driver->err_handler ||
249 !dev->driver->err_handler->error_detected) {
250 if (result_data->state == pci_channel_io_frozen &&
251 dev->hdr_type != PCI_HEADER_TYPE_BRIDGE) {
253 * In case of fatal recovery, if one of down-
254 * stream device has no driver. We might be
255 * unable to recover because a later insmod
256 * of a driver for this device is unaware of
259 dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
261 "no AER-aware driver" : "no driver");
265 * If there's any device in the subtree that does not
266 * have an error_detected callback, returning
267 * PCI_ERS_RESULT_NO_AER_DRIVER prevents calling of
268 * the subsequent mmio_enabled/slot_reset/resume
269 * callbacks of "any" device in the subtree. All the
270 * devices in the subtree are left in the error state
274 if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE)
275 vote = PCI_ERS_RESULT_NO_AER_DRIVER;
277 vote = PCI_ERS_RESULT_NONE;
279 err_handler = dev->driver->err_handler;
280 vote = err_handler->error_detected(dev, result_data->state);
281 pci_uevent_ers(dev, PCI_ERS_RESULT_NONE);
284 result_data->result = merge_result(result_data->result, vote);
285 device_unlock(&dev->dev);
289 static int report_mmio_enabled(struct pci_dev *dev, void *data)
291 pci_ers_result_t vote;
292 const struct pci_error_handlers *err_handler;
293 struct aer_broadcast_data *result_data;
294 result_data = (struct aer_broadcast_data *) data;
296 device_lock(&dev->dev);
298 !dev->driver->err_handler ||
299 !dev->driver->err_handler->mmio_enabled)
302 err_handler = dev->driver->err_handler;
303 vote = err_handler->mmio_enabled(dev);
304 result_data->result = merge_result(result_data->result, vote);
306 device_unlock(&dev->dev);
310 static int report_slot_reset(struct pci_dev *dev, void *data)
312 pci_ers_result_t vote;
313 const struct pci_error_handlers *err_handler;
314 struct aer_broadcast_data *result_data;
315 result_data = (struct aer_broadcast_data *) data;
317 device_lock(&dev->dev);
319 !dev->driver->err_handler ||
320 !dev->driver->err_handler->slot_reset)
323 err_handler = dev->driver->err_handler;
324 vote = err_handler->slot_reset(dev);
325 result_data->result = merge_result(result_data->result, vote);
327 device_unlock(&dev->dev);
331 static int report_resume(struct pci_dev *dev, void *data)
333 const struct pci_error_handlers *err_handler;
335 device_lock(&dev->dev);
336 dev->error_state = pci_channel_io_normal;
339 !dev->driver->err_handler ||
340 !dev->driver->err_handler->resume)
343 err_handler = dev->driver->err_handler;
344 err_handler->resume(dev);
345 pci_uevent_ers(dev, PCI_ERS_RESULT_RECOVERED);
347 device_unlock(&dev->dev);
352 * broadcast_error_message - handle message broadcast to downstream drivers
353 * @dev: pointer to from where in a hierarchy message is broadcasted down
354 * @state: error state
355 * @error_mesg: message to print
356 * @cb: callback to be broadcasted
358 * Invoked during error recovery process. Once being invoked, the content
359 * of error severity will be broadcasted to all downstream drivers in a
360 * hierarchy in question.
362 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
363 enum pci_channel_state state,
365 int (*cb)(struct pci_dev *, void *))
367 struct aer_broadcast_data result_data;
369 dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
370 result_data.state = state;
371 if (cb == report_error_detected)
372 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
374 result_data.result = PCI_ERS_RESULT_RECOVERED;
376 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
378 * If the error is reported by a bridge, we think this error
379 * is related to the downstream link of the bridge, so we
380 * do error recovery on all subordinates of the bridge instead
381 * of the bridge and clear the error status of the bridge.
383 if (cb == report_error_detected)
384 dev->error_state = state;
385 pci_walk_bus(dev->subordinate, cb, &result_data);
386 if (cb == report_resume) {
387 pci_cleanup_aer_uncorrect_error_status(dev);
388 dev->error_state = pci_channel_io_normal;
392 * If the error is reported by an end point, we think this
393 * error is related to the upstream link of the end point.
395 if (state == pci_channel_io_normal)
397 * the error is non fatal so the bus is ok, just invoke
398 * the callback for the function that logged the error.
400 cb(dev, &result_data);
402 pci_walk_bus(dev->bus, cb, &result_data);
405 return result_data.result;
409 * default_reset_link - default reset function
410 * @dev: pointer to pci_dev data structure
412 * Invoked when performing link reset on a Downstream Port or a
413 * Root Port with no aer driver.
415 static pci_ers_result_t default_reset_link(struct pci_dev *dev)
417 pci_reset_bridge_secondary_bus(dev);
418 dev_printk(KERN_DEBUG, &dev->dev, "downstream link has been reset\n");
419 return PCI_ERS_RESULT_RECOVERED;
422 static int find_aer_service_iter(struct device *device, void *data)
424 struct pcie_port_service_driver *service_driver, **drv;
426 drv = (struct pcie_port_service_driver **) data;
428 if (device->bus == &pcie_port_bus_type && device->driver) {
429 service_driver = to_service_driver(device->driver);
430 if (service_driver->service == PCIE_PORT_SERVICE_AER) {
431 *drv = service_driver;
439 static struct pcie_port_service_driver *find_aer_service(struct pci_dev *dev)
441 struct pcie_port_service_driver *drv = NULL;
443 device_for_each_child(&dev->dev, &drv, find_aer_service_iter);
448 static pci_ers_result_t reset_link(struct pci_dev *dev)
450 struct pci_dev *udev;
451 pci_ers_result_t status;
452 struct pcie_port_service_driver *driver;
454 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
455 /* Reset this port for all subordinates */
458 /* Reset the upstream component (likely downstream port) */
459 udev = dev->bus->self;
462 /* Use the aer driver of the component firstly */
463 driver = find_aer_service(udev);
465 if (driver && driver->reset_link) {
466 status = driver->reset_link(udev);
467 } else if (udev->has_secondary_link) {
468 status = default_reset_link(udev);
470 dev_printk(KERN_DEBUG, &dev->dev,
471 "no link-reset support at upstream device %s\n",
473 return PCI_ERS_RESULT_DISCONNECT;
476 if (status != PCI_ERS_RESULT_RECOVERED) {
477 dev_printk(KERN_DEBUG, &dev->dev,
478 "link reset at upstream device %s failed\n",
480 return PCI_ERS_RESULT_DISCONNECT;
487 * do_recovery - handle nonfatal/fatal error recovery process
488 * @dev: pointer to a pci_dev data structure of agent detecting an error
489 * @severity: error severity type
491 * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
492 * error detected message to all downstream drivers within a hierarchy in
493 * question and return the returned code.
495 static void do_recovery(struct pci_dev *dev, int severity)
497 pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
498 enum pci_channel_state state;
500 if (severity == AER_FATAL)
501 state = pci_channel_io_frozen;
503 state = pci_channel_io_normal;
505 status = broadcast_error_message(dev,
508 report_error_detected);
510 if (severity == AER_FATAL) {
511 result = reset_link(dev);
512 if (result != PCI_ERS_RESULT_RECOVERED)
516 if (status == PCI_ERS_RESULT_CAN_RECOVER)
517 status = broadcast_error_message(dev,
520 report_mmio_enabled);
522 if (status == PCI_ERS_RESULT_NEED_RESET) {
524 * TODO: Should call platform-specific
525 * functions to reset slot before calling
526 * drivers' slot_reset callbacks?
528 status = broadcast_error_message(dev,
534 if (status != PCI_ERS_RESULT_RECOVERED)
537 broadcast_error_message(dev,
542 dev_info(&dev->dev, "AER: Device recovery successful\n");
546 pci_uevent_ers(dev, PCI_ERS_RESULT_DISCONNECT);
547 /* TODO: Should kernel panic here? */
548 dev_info(&dev->dev, "AER: Device recovery failed\n");
552 * handle_error_source - handle logging error into an event log
553 * @aerdev: pointer to pcie_device data structure of the root port
554 * @dev: pointer to pci_dev data structure of error source device
555 * @info: comprehensive error information
557 * Invoked when an error being detected by Root Port.
559 static void handle_error_source(struct pcie_device *aerdev,
561 struct aer_err_info *info)
565 if (info->severity == AER_CORRECTABLE) {
567 * Correctable error does not need software intervention.
568 * No need to go through error recovery process.
572 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
575 do_recovery(dev, info->severity);
578 #ifdef CONFIG_ACPI_APEI_PCIEAER
579 static void aer_recover_work_func(struct work_struct *work);
581 #define AER_RECOVER_RING_ORDER 4
582 #define AER_RECOVER_RING_SIZE (1 << AER_RECOVER_RING_ORDER)
584 struct aer_recover_entry {
589 struct aer_capability_regs *regs;
592 static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
593 AER_RECOVER_RING_SIZE);
595 * Mutual exclusion for writers of aer_recover_ring, reader side don't
596 * need lock, because there is only one reader and lock is not needed
597 * between reader and writer.
599 static DEFINE_SPINLOCK(aer_recover_ring_lock);
600 static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
602 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
603 int severity, struct aer_capability_regs *aer_regs)
606 struct aer_recover_entry entry = {
610 .severity = severity,
614 spin_lock_irqsave(&aer_recover_ring_lock, flags);
615 if (kfifo_put(&aer_recover_ring, entry))
616 schedule_work(&aer_recover_work);
618 pr_err("AER recover: Buffer overflow when recovering AER for %04x:%02x:%02x:%x\n",
619 domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
620 spin_unlock_irqrestore(&aer_recover_ring_lock, flags);
622 EXPORT_SYMBOL_GPL(aer_recover_queue);
624 static void aer_recover_work_func(struct work_struct *work)
626 struct aer_recover_entry entry;
627 struct pci_dev *pdev;
629 while (kfifo_get(&aer_recover_ring, &entry)) {
630 pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
633 pr_err("AER recover: Can not find pci_dev for %04x:%02x:%02x:%x\n",
634 entry.domain, entry.bus,
635 PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
638 cper_print_aer(pdev, entry.severity, entry.regs);
639 do_recovery(pdev, entry.severity);
646 * get_device_error_info - read error status from dev and store it to info
647 * @dev: pointer to the device expected to have a error record
648 * @info: pointer to structure to store the error record
650 * Return 1 on success, 0 on error.
652 * Note that @info is reused among all error devices. Clear fields properly.
654 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
658 /* Must reset in this function */
660 info->tlp_header_valid = 0;
664 /* The device might not support AER */
668 if (info->severity == AER_CORRECTABLE) {
669 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
671 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
673 if (!(info->status & ~info->mask))
675 } else if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
676 info->severity == AER_NONFATAL) {
678 /* Link is still healthy for IO reads */
679 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
681 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
683 if (!(info->status & ~info->mask))
686 /* Get First Error Pointer */
687 pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
688 info->first_error = PCI_ERR_CAP_FEP(temp);
690 if (info->status & AER_LOG_TLP_MASKS) {
691 info->tlp_header_valid = 1;
692 pci_read_config_dword(dev,
693 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
694 pci_read_config_dword(dev,
695 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
696 pci_read_config_dword(dev,
697 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
698 pci_read_config_dword(dev,
699 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
706 static inline void aer_process_err_devices(struct pcie_device *p_device,
707 struct aer_err_info *e_info)
711 /* Report all before handle them, not to lost records by reset etc. */
712 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
713 if (get_device_error_info(e_info->dev[i], e_info))
714 aer_print_error(e_info->dev[i], e_info);
716 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
717 if (get_device_error_info(e_info->dev[i], e_info))
718 handle_error_source(p_device, e_info->dev[i], e_info);
723 * aer_isr_one_error - consume an error detected by root port
724 * @p_device: pointer to error root port service device
725 * @e_src: pointer to an error source
727 static void aer_isr_one_error(struct pcie_device *p_device,
728 struct aer_err_source *e_src)
730 struct aer_rpc *rpc = get_service_data(p_device);
731 struct aer_err_info *e_info = &rpc->e_info;
734 * There is a possibility that both correctable error and
735 * uncorrectable error being logged. Report correctable error first.
737 if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
738 e_info->id = ERR_COR_ID(e_src->id);
739 e_info->severity = AER_CORRECTABLE;
741 if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
742 e_info->multi_error_valid = 1;
744 e_info->multi_error_valid = 0;
746 aer_print_port_info(p_device->port, e_info);
748 if (find_source_device(p_device->port, e_info))
749 aer_process_err_devices(p_device, e_info);
752 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
753 e_info->id = ERR_UNCOR_ID(e_src->id);
755 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
756 e_info->severity = AER_FATAL;
758 e_info->severity = AER_NONFATAL;
760 if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
761 e_info->multi_error_valid = 1;
763 e_info->multi_error_valid = 0;
765 aer_print_port_info(p_device->port, e_info);
767 if (find_source_device(p_device->port, e_info))
768 aer_process_err_devices(p_device, e_info);
773 * get_e_source - retrieve an error source
774 * @rpc: pointer to the root port which holds an error
775 * @e_src: pointer to store retrieved error source
777 * Return 1 if an error source is retrieved, otherwise 0.
779 * Invoked by DPC handler to consume an error.
781 static int get_e_source(struct aer_rpc *rpc, struct aer_err_source *e_src)
785 /* Lock access to Root error producer/consumer index */
786 spin_lock_irqsave(&rpc->e_lock, flags);
787 if (rpc->prod_idx == rpc->cons_idx) {
788 spin_unlock_irqrestore(&rpc->e_lock, flags);
792 *e_src = rpc->e_sources[rpc->cons_idx];
794 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
796 spin_unlock_irqrestore(&rpc->e_lock, flags);
802 * aer_isr - consume errors detected by root port
803 * @work: definition of this work item
805 * Invoked, as DPC, when root port records new detected error
807 void aer_isr(struct work_struct *work)
809 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
810 struct pcie_device *p_device = rpc->rpd;
811 struct aer_err_source uninitialized_var(e_src);
813 mutex_lock(&rpc->rpc_mutex);
814 while (get_e_source(rpc, &e_src))
815 aer_isr_one_error(p_device, &e_src);
816 mutex_unlock(&rpc->rpc_mutex);