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>
26 #include <linux/slab.h>
27 #include <linux/kfifo.h>
30 static bool forceload;
31 static bool nosourceid;
32 module_param(forceload, bool, 0);
33 module_param(nosourceid, bool, 0);
35 #define PCI_EXP_AER_FLAGS (PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \
36 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE)
38 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
40 if (pcie_aer_get_firmware_first(dev))
43 if (!pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR))
46 return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
48 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
50 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
52 if (pcie_aer_get_firmware_first(dev))
55 return pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
58 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
60 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
65 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
69 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
71 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
75 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
78 * add_error_device - list device to be handled
79 * @e_info: pointer to error info
80 * @dev: pointer to pci_dev to be added
82 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
84 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
85 e_info->dev[e_info->error_dev_num] = dev;
86 e_info->error_dev_num++;
92 #define PCI_BUS(x) (((x) >> 8) & 0xff)
95 * is_error_source - check whether the device is source of reported error
96 * @dev: pointer to pci_dev to be checked
97 * @e_info: pointer to reported error info
99 static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info)
106 * When bus id is equal to 0, it might be a bad id
107 * reported by root port.
109 if (!nosourceid && (PCI_BUS(e_info->id) != 0)) {
110 /* Device ID match? */
111 if (e_info->id == ((dev->bus->number << 8) | dev->devfn))
114 /* Continue id comparing if there is no multiple error */
115 if (!e_info->multi_error_valid)
122 * 2) bus id is equal to 0. Some ports might lose the bus
123 * id of error source id;
124 * 3) There are multiple errors and prior id comparing fails;
125 * We check AER status registers to find possible reporter.
127 if (atomic_read(&dev->enable_cnt) == 0)
130 /* Check if AER is enabled */
131 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, ®16);
132 if (!(reg16 & PCI_EXP_AER_FLAGS))
135 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
139 /* Check if error is recorded */
140 if (e_info->severity == AER_CORRECTABLE) {
141 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
142 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, &mask);
144 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
145 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
153 static int find_device_iter(struct pci_dev *dev, void *data)
155 struct aer_err_info *e_info = (struct aer_err_info *)data;
157 if (is_error_source(dev, e_info)) {
158 /* List this device */
159 if (add_error_device(e_info, dev)) {
160 /* We cannot handle more... Stop iteration */
161 /* TODO: Should print error message here? */
165 /* If there is only a single error, stop iteration */
166 if (!e_info->multi_error_valid)
173 * find_source_device - search through device hierarchy for source device
174 * @parent: pointer to Root Port pci_dev data structure
175 * @e_info: including detailed error information such like id
177 * Return true if found.
179 * Invoked by DPC when error is detected at the Root Port.
180 * Caller of this function must set id, severity, and multi_error_valid of
181 * struct aer_err_info pointed by @e_info properly. This function must fill
182 * e_info->error_dev_num and e_info->dev[], based on the given information.
184 static bool find_source_device(struct pci_dev *parent,
185 struct aer_err_info *e_info)
187 struct pci_dev *dev = parent;
190 /* Must reset in this function */
191 e_info->error_dev_num = 0;
193 /* Is Root Port an agent that sends error message? */
194 result = find_device_iter(dev, e_info);
198 pci_walk_bus(parent->subordinate, find_device_iter, e_info);
200 if (!e_info->error_dev_num) {
201 dev_printk(KERN_DEBUG, &parent->dev,
202 "can't find device of ID%04x\n",
209 static int report_error_detected(struct pci_dev *dev, void *data)
211 pci_ers_result_t vote;
212 const struct pci_error_handlers *err_handler;
213 struct aer_broadcast_data *result_data;
214 result_data = (struct aer_broadcast_data *) data;
216 device_lock(&dev->dev);
217 dev->error_state = result_data->state;
220 !dev->driver->err_handler ||
221 !dev->driver->err_handler->error_detected) {
222 if (result_data->state == pci_channel_io_frozen &&
223 !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
225 * In case of fatal recovery, if one of down-
226 * stream device has no driver. We might be
227 * unable to recover because a later insmod
228 * of a driver for this device is unaware of
231 dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
233 "no AER-aware driver" : "no driver");
238 err_handler = dev->driver->err_handler;
239 vote = err_handler->error_detected(dev, result_data->state);
240 result_data->result = merge_result(result_data->result, vote);
242 device_unlock(&dev->dev);
246 static int report_mmio_enabled(struct pci_dev *dev, void *data)
248 pci_ers_result_t vote;
249 const struct pci_error_handlers *err_handler;
250 struct aer_broadcast_data *result_data;
251 result_data = (struct aer_broadcast_data *) data;
253 device_lock(&dev->dev);
255 !dev->driver->err_handler ||
256 !dev->driver->err_handler->mmio_enabled)
259 err_handler = dev->driver->err_handler;
260 vote = err_handler->mmio_enabled(dev);
261 result_data->result = merge_result(result_data->result, vote);
263 device_unlock(&dev->dev);
267 static int report_slot_reset(struct pci_dev *dev, void *data)
269 pci_ers_result_t vote;
270 const struct pci_error_handlers *err_handler;
271 struct aer_broadcast_data *result_data;
272 result_data = (struct aer_broadcast_data *) data;
274 device_lock(&dev->dev);
276 !dev->driver->err_handler ||
277 !dev->driver->err_handler->slot_reset)
280 err_handler = dev->driver->err_handler;
281 vote = err_handler->slot_reset(dev);
282 result_data->result = merge_result(result_data->result, vote);
284 device_unlock(&dev->dev);
288 static int report_resume(struct pci_dev *dev, void *data)
290 const struct pci_error_handlers *err_handler;
292 device_lock(&dev->dev);
293 dev->error_state = pci_channel_io_normal;
296 !dev->driver->err_handler ||
297 !dev->driver->err_handler->resume)
300 err_handler = dev->driver->err_handler;
301 err_handler->resume(dev);
303 device_unlock(&dev->dev);
308 * broadcast_error_message - handle message broadcast to downstream drivers
309 * @dev: pointer to from where in a hierarchy message is broadcasted down
310 * @state: error state
311 * @error_mesg: message to print
312 * @cb: callback to be broadcasted
314 * Invoked during error recovery process. Once being invoked, the content
315 * of error severity will be broadcasted to all downstream drivers in a
316 * hierarchy in question.
318 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
319 enum pci_channel_state state,
321 int (*cb)(struct pci_dev *, void *))
323 struct aer_broadcast_data result_data;
325 dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
326 result_data.state = state;
327 if (cb == report_error_detected)
328 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
330 result_data.result = PCI_ERS_RESULT_RECOVERED;
332 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
334 * If the error is reported by a bridge, we think this error
335 * is related to the downstream link of the bridge, so we
336 * do error recovery on all subordinates of the bridge instead
337 * of the bridge and clear the error status of the bridge.
339 if (cb == report_error_detected)
340 dev->error_state = state;
341 pci_walk_bus(dev->subordinate, cb, &result_data);
342 if (cb == report_resume) {
343 pci_cleanup_aer_uncorrect_error_status(dev);
344 dev->error_state = pci_channel_io_normal;
348 * If the error is reported by an end point, we think this
349 * error is related to the upstream link of the end point.
351 pci_walk_bus(dev->bus, cb, &result_data);
354 return result_data.result;
358 * aer_do_secondary_bus_reset - perform secondary bus reset
359 * @dev: pointer to bridge's pci_dev data structure
361 * Invoked when performing link reset at Root Port or Downstream Port.
363 void aer_do_secondary_bus_reset(struct pci_dev *dev)
367 /* Assert Secondary Bus Reset */
368 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &p2p_ctrl);
369 p2p_ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
370 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, p2p_ctrl);
373 * we should send hot reset message for 2ms to allow it time to
374 * propagate to all downstream ports
378 /* De-assert Secondary Bus Reset */
379 p2p_ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
380 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, p2p_ctrl);
383 * System software must wait for at least 100ms from the end
384 * of a reset of one or more device before it is permitted
385 * to issue Configuration Requests to those devices.
391 * default_downstream_reset_link - default reset function for Downstream Port
392 * @dev: pointer to downstream port's pci_dev data structure
394 * Invoked when performing link reset at Downstream Port w/ no aer driver.
396 static pci_ers_result_t default_downstream_reset_link(struct pci_dev *dev)
398 aer_do_secondary_bus_reset(dev);
399 dev_printk(KERN_DEBUG, &dev->dev,
400 "Downstream Port link has been reset\n");
401 return PCI_ERS_RESULT_RECOVERED;
404 static int find_aer_service_iter(struct device *device, void *data)
406 struct pcie_port_service_driver *service_driver, **drv;
408 drv = (struct pcie_port_service_driver **) data;
410 if (device->bus == &pcie_port_bus_type && device->driver) {
411 service_driver = to_service_driver(device->driver);
412 if (service_driver->service == PCIE_PORT_SERVICE_AER) {
413 *drv = service_driver;
421 static struct pcie_port_service_driver *find_aer_service(struct pci_dev *dev)
423 struct pcie_port_service_driver *drv = NULL;
425 device_for_each_child(&dev->dev, &drv, find_aer_service_iter);
430 static pci_ers_result_t reset_link(struct pci_dev *dev)
432 struct pci_dev *udev;
433 pci_ers_result_t status;
434 struct pcie_port_service_driver *driver;
436 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
437 /* Reset this port for all subordinates */
440 /* Reset the upstream component (likely downstream port) */
441 udev = dev->bus->self;
444 /* Use the aer driver of the component firstly */
445 driver = find_aer_service(udev);
447 if (driver && driver->reset_link) {
448 status = driver->reset_link(udev);
449 } else if (pci_pcie_type(udev) == PCI_EXP_TYPE_DOWNSTREAM) {
450 status = default_downstream_reset_link(udev);
452 dev_printk(KERN_DEBUG, &dev->dev,
453 "no link-reset support at upstream device %s\n",
455 return PCI_ERS_RESULT_DISCONNECT;
458 if (status != PCI_ERS_RESULT_RECOVERED) {
459 dev_printk(KERN_DEBUG, &dev->dev,
460 "link reset at upstream device %s failed\n",
462 return PCI_ERS_RESULT_DISCONNECT;
469 * do_recovery - handle nonfatal/fatal error recovery process
470 * @dev: pointer to a pci_dev data structure of agent detecting an error
471 * @severity: error severity type
473 * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
474 * error detected message to all downstream drivers within a hierarchy in
475 * question and return the returned code.
477 static void do_recovery(struct pci_dev *dev, int severity)
479 pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
480 enum pci_channel_state state;
482 if (severity == AER_FATAL)
483 state = pci_channel_io_frozen;
485 state = pci_channel_io_normal;
487 status = broadcast_error_message(dev,
490 report_error_detected);
492 if (severity == AER_FATAL) {
493 result = reset_link(dev);
494 if (result != PCI_ERS_RESULT_RECOVERED)
498 if (status == PCI_ERS_RESULT_CAN_RECOVER)
499 status = broadcast_error_message(dev,
502 report_mmio_enabled);
504 if (status == PCI_ERS_RESULT_NEED_RESET) {
506 * TODO: Should call platform-specific
507 * functions to reset slot before calling
508 * drivers' slot_reset callbacks?
510 status = broadcast_error_message(dev,
516 if (status != PCI_ERS_RESULT_RECOVERED)
519 broadcast_error_message(dev,
524 dev_info(&dev->dev, "AER: Device recovery successful\n");
528 /* TODO: Should kernel panic here? */
529 dev_info(&dev->dev, "AER: Device recovery failed\n");
533 * handle_error_source - handle logging error into an event log
534 * @aerdev: pointer to pcie_device data structure of the root port
535 * @dev: pointer to pci_dev data structure of error source device
536 * @info: comprehensive error information
538 * Invoked when an error being detected by Root Port.
540 static void handle_error_source(struct pcie_device *aerdev,
542 struct aer_err_info *info)
546 if (info->severity == AER_CORRECTABLE) {
548 * Correctable error does not need software intevention.
549 * No need to go through error recovery process.
551 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
553 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
556 do_recovery(dev, info->severity);
559 #ifdef CONFIG_ACPI_APEI_PCIEAER
560 static void aer_recover_work_func(struct work_struct *work);
562 #define AER_RECOVER_RING_ORDER 4
563 #define AER_RECOVER_RING_SIZE (1 << AER_RECOVER_RING_ORDER)
565 struct aer_recover_entry
573 static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
574 AER_RECOVER_RING_SIZE);
576 * Mutual exclusion for writers of aer_recover_ring, reader side don't
577 * need lock, because there is only one reader and lock is not needed
578 * between reader and writer.
580 static DEFINE_SPINLOCK(aer_recover_ring_lock);
581 static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
583 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
587 struct aer_recover_entry entry = {
591 .severity = severity,
594 spin_lock_irqsave(&aer_recover_ring_lock, flags);
595 if (kfifo_put(&aer_recover_ring, &entry))
596 schedule_work(&aer_recover_work);
598 pr_err("AER recover: Buffer overflow when recovering AER for %04x:%02x:%02x:%x\n",
599 domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
600 spin_unlock_irqrestore(&aer_recover_ring_lock, flags);
602 EXPORT_SYMBOL_GPL(aer_recover_queue);
604 static void aer_recover_work_func(struct work_struct *work)
606 struct aer_recover_entry entry;
607 struct pci_dev *pdev;
609 while (kfifo_get(&aer_recover_ring, &entry)) {
610 pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
613 pr_err("AER recover: Can not find pci_dev for %04x:%02x:%02x:%x\n",
614 entry.domain, entry.bus,
615 PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
618 do_recovery(pdev, entry.severity);
624 * get_device_error_info - read error status from dev and store it to info
625 * @dev: pointer to the device expected to have a error record
626 * @info: pointer to structure to store the error record
628 * Return 1 on success, 0 on error.
630 * Note that @info is reused among all error devices. Clear fields properly.
632 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
636 /* Must reset in this function */
638 info->tlp_header_valid = 0;
640 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
642 /* The device might not support AER */
646 if (info->severity == AER_CORRECTABLE) {
647 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
649 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
651 if (!(info->status & ~info->mask))
653 } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
654 info->severity == AER_NONFATAL) {
656 /* Link is still healthy for IO reads */
657 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
659 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
661 if (!(info->status & ~info->mask))
664 /* Get First Error Pointer */
665 pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
666 info->first_error = PCI_ERR_CAP_FEP(temp);
668 if (info->status & AER_LOG_TLP_MASKS) {
669 info->tlp_header_valid = 1;
670 pci_read_config_dword(dev,
671 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
672 pci_read_config_dword(dev,
673 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
674 pci_read_config_dword(dev,
675 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
676 pci_read_config_dword(dev,
677 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
684 static inline void aer_process_err_devices(struct pcie_device *p_device,
685 struct aer_err_info *e_info)
689 /* Report all before handle them, not to lost records by reset etc. */
690 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
691 if (get_device_error_info(e_info->dev[i], e_info))
692 aer_print_error(e_info->dev[i], e_info);
694 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
695 if (get_device_error_info(e_info->dev[i], e_info))
696 handle_error_source(p_device, e_info->dev[i], e_info);
701 * aer_isr_one_error - consume an error detected by root port
702 * @p_device: pointer to error root port service device
703 * @e_src: pointer to an error source
705 static void aer_isr_one_error(struct pcie_device *p_device,
706 struct aer_err_source *e_src)
708 struct aer_err_info *e_info;
710 /* struct aer_err_info might be big, so we allocate it with slab */
711 e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL);
713 dev_printk(KERN_DEBUG, &p_device->port->dev,
714 "Can't allocate mem when processing AER errors\n");
719 * There is a possibility that both correctable error and
720 * uncorrectable error being logged. Report correctable error first.
722 if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
723 e_info->id = ERR_COR_ID(e_src->id);
724 e_info->severity = AER_CORRECTABLE;
726 if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
727 e_info->multi_error_valid = 1;
729 e_info->multi_error_valid = 0;
731 aer_print_port_info(p_device->port, e_info);
733 if (find_source_device(p_device->port, e_info))
734 aer_process_err_devices(p_device, e_info);
737 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
738 e_info->id = ERR_UNCOR_ID(e_src->id);
740 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
741 e_info->severity = AER_FATAL;
743 e_info->severity = AER_NONFATAL;
745 if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
746 e_info->multi_error_valid = 1;
748 e_info->multi_error_valid = 0;
750 aer_print_port_info(p_device->port, e_info);
752 if (find_source_device(p_device->port, e_info))
753 aer_process_err_devices(p_device, e_info);
760 * get_e_source - retrieve an error source
761 * @rpc: pointer to the root port which holds an error
762 * @e_src: pointer to store retrieved error source
764 * Return 1 if an error source is retrieved, otherwise 0.
766 * Invoked by DPC handler to consume an error.
768 static int get_e_source(struct aer_rpc *rpc, struct aer_err_source *e_src)
772 /* Lock access to Root error producer/consumer index */
773 spin_lock_irqsave(&rpc->e_lock, flags);
774 if (rpc->prod_idx == rpc->cons_idx) {
775 spin_unlock_irqrestore(&rpc->e_lock, flags);
779 *e_src = rpc->e_sources[rpc->cons_idx];
781 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
783 spin_unlock_irqrestore(&rpc->e_lock, flags);
789 * aer_isr - consume errors detected by root port
790 * @work: definition of this work item
792 * Invoked, as DPC, when root port records new detected error
794 void aer_isr(struct work_struct *work)
796 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
797 struct pcie_device *p_device = rpc->rpd;
798 struct aer_err_source uninitialized_var(e_src);
800 mutex_lock(&rpc->rpc_mutex);
801 while (get_e_source(rpc, &e_src))
802 aer_isr_one_error(p_device, &e_src);
803 mutex_unlock(&rpc->rpc_mutex);
805 wake_up(&rpc->wait_release);
809 * aer_init - provide AER initialization
810 * @dev: pointer to AER pcie device
812 * Invoked when AER service driver is loaded.
814 int aer_init(struct pcie_device *dev)
817 dev_printk(KERN_DEBUG, &dev->device,
818 "aerdrv forceload requested.\n");
819 pcie_aer_force_firmware_first(dev->port, 0);