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");
237 * If there's any device in the subtree that does not
238 * have an error_detected callback, returning
239 * PCI_ERS_RESULT_NO_AER_DRIVER prevents calling of
240 * the subsequent mmio_enabled/slot_reset/resume
241 * callbacks of "any" device in the subtree. All the
242 * devices in the subtree are left in the error state
246 if (!(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
247 vote = PCI_ERS_RESULT_NO_AER_DRIVER;
249 vote = PCI_ERS_RESULT_NONE;
251 err_handler = dev->driver->err_handler;
252 vote = err_handler->error_detected(dev, result_data->state);
255 result_data->result = merge_result(result_data->result, vote);
256 device_unlock(&dev->dev);
260 static int report_mmio_enabled(struct pci_dev *dev, void *data)
262 pci_ers_result_t vote;
263 const struct pci_error_handlers *err_handler;
264 struct aer_broadcast_data *result_data;
265 result_data = (struct aer_broadcast_data *) data;
267 device_lock(&dev->dev);
269 !dev->driver->err_handler ||
270 !dev->driver->err_handler->mmio_enabled)
273 err_handler = dev->driver->err_handler;
274 vote = err_handler->mmio_enabled(dev);
275 result_data->result = merge_result(result_data->result, vote);
277 device_unlock(&dev->dev);
281 static int report_slot_reset(struct pci_dev *dev, void *data)
283 pci_ers_result_t vote;
284 const struct pci_error_handlers *err_handler;
285 struct aer_broadcast_data *result_data;
286 result_data = (struct aer_broadcast_data *) data;
288 device_lock(&dev->dev);
290 !dev->driver->err_handler ||
291 !dev->driver->err_handler->slot_reset)
294 err_handler = dev->driver->err_handler;
295 vote = err_handler->slot_reset(dev);
296 result_data->result = merge_result(result_data->result, vote);
298 device_unlock(&dev->dev);
302 static int report_resume(struct pci_dev *dev, void *data)
304 const struct pci_error_handlers *err_handler;
306 device_lock(&dev->dev);
307 dev->error_state = pci_channel_io_normal;
310 !dev->driver->err_handler ||
311 !dev->driver->err_handler->resume)
314 err_handler = dev->driver->err_handler;
315 err_handler->resume(dev);
317 device_unlock(&dev->dev);
322 * broadcast_error_message - handle message broadcast to downstream drivers
323 * @dev: pointer to from where in a hierarchy message is broadcasted down
324 * @state: error state
325 * @error_mesg: message to print
326 * @cb: callback to be broadcasted
328 * Invoked during error recovery process. Once being invoked, the content
329 * of error severity will be broadcasted to all downstream drivers in a
330 * hierarchy in question.
332 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
333 enum pci_channel_state state,
335 int (*cb)(struct pci_dev *, void *))
337 struct aer_broadcast_data result_data;
339 dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
340 result_data.state = state;
341 if (cb == report_error_detected)
342 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
344 result_data.result = PCI_ERS_RESULT_RECOVERED;
346 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
348 * If the error is reported by a bridge, we think this error
349 * is related to the downstream link of the bridge, so we
350 * do error recovery on all subordinates of the bridge instead
351 * of the bridge and clear the error status of the bridge.
353 if (cb == report_error_detected)
354 dev->error_state = state;
355 pci_walk_bus(dev->subordinate, cb, &result_data);
356 if (cb == report_resume) {
357 pci_cleanup_aer_uncorrect_error_status(dev);
358 dev->error_state = pci_channel_io_normal;
362 * If the error is reported by an end point, we think this
363 * error is related to the upstream link of the end point.
365 pci_walk_bus(dev->bus, cb, &result_data);
368 return result_data.result;
372 * aer_do_secondary_bus_reset - perform secondary bus reset
373 * @dev: pointer to bridge's pci_dev data structure
375 * Invoked when performing link reset at Root Port or Downstream Port.
377 void aer_do_secondary_bus_reset(struct pci_dev *dev)
381 /* Assert Secondary Bus Reset */
382 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &p2p_ctrl);
383 p2p_ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
384 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, p2p_ctrl);
387 * we should send hot reset message for 2ms to allow it time to
388 * propagate to all downstream ports
392 /* De-assert Secondary Bus Reset */
393 p2p_ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
394 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, p2p_ctrl);
397 * System software must wait for at least 100ms from the end
398 * of a reset of one or more device before it is permitted
399 * to issue Configuration Requests to those devices.
405 * default_downstream_reset_link - default reset function for Downstream Port
406 * @dev: pointer to downstream port's pci_dev data structure
408 * Invoked when performing link reset at Downstream Port w/ no aer driver.
410 static pci_ers_result_t default_downstream_reset_link(struct pci_dev *dev)
412 aer_do_secondary_bus_reset(dev);
413 dev_printk(KERN_DEBUG, &dev->dev,
414 "Downstream Port link has been reset\n");
415 return PCI_ERS_RESULT_RECOVERED;
418 static int find_aer_service_iter(struct device *device, void *data)
420 struct pcie_port_service_driver *service_driver, **drv;
422 drv = (struct pcie_port_service_driver **) data;
424 if (device->bus == &pcie_port_bus_type && device->driver) {
425 service_driver = to_service_driver(device->driver);
426 if (service_driver->service == PCIE_PORT_SERVICE_AER) {
427 *drv = service_driver;
435 static struct pcie_port_service_driver *find_aer_service(struct pci_dev *dev)
437 struct pcie_port_service_driver *drv = NULL;
439 device_for_each_child(&dev->dev, &drv, find_aer_service_iter);
444 static pci_ers_result_t reset_link(struct pci_dev *dev)
446 struct pci_dev *udev;
447 pci_ers_result_t status;
448 struct pcie_port_service_driver *driver;
450 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
451 /* Reset this port for all subordinates */
454 /* Reset the upstream component (likely downstream port) */
455 udev = dev->bus->self;
458 /* Use the aer driver of the component firstly */
459 driver = find_aer_service(udev);
461 if (driver && driver->reset_link) {
462 status = driver->reset_link(udev);
463 } else if (pci_pcie_type(udev) == PCI_EXP_TYPE_DOWNSTREAM) {
464 status = default_downstream_reset_link(udev);
466 dev_printk(KERN_DEBUG, &dev->dev,
467 "no link-reset support at upstream device %s\n",
469 return PCI_ERS_RESULT_DISCONNECT;
472 if (status != PCI_ERS_RESULT_RECOVERED) {
473 dev_printk(KERN_DEBUG, &dev->dev,
474 "link reset at upstream device %s failed\n",
476 return PCI_ERS_RESULT_DISCONNECT;
483 * do_recovery - handle nonfatal/fatal error recovery process
484 * @dev: pointer to a pci_dev data structure of agent detecting an error
485 * @severity: error severity type
487 * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
488 * error detected message to all downstream drivers within a hierarchy in
489 * question and return the returned code.
491 static void do_recovery(struct pci_dev *dev, int severity)
493 pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
494 enum pci_channel_state state;
496 if (severity == AER_FATAL)
497 state = pci_channel_io_frozen;
499 state = pci_channel_io_normal;
501 status = broadcast_error_message(dev,
504 report_error_detected);
506 if (severity == AER_FATAL) {
507 result = reset_link(dev);
508 if (result != PCI_ERS_RESULT_RECOVERED)
512 if (status == PCI_ERS_RESULT_CAN_RECOVER)
513 status = broadcast_error_message(dev,
516 report_mmio_enabled);
518 if (status == PCI_ERS_RESULT_NEED_RESET) {
520 * TODO: Should call platform-specific
521 * functions to reset slot before calling
522 * drivers' slot_reset callbacks?
524 status = broadcast_error_message(dev,
530 if (status != PCI_ERS_RESULT_RECOVERED)
533 broadcast_error_message(dev,
538 dev_info(&dev->dev, "AER: Device recovery successful\n");
542 /* TODO: Should kernel panic here? */
543 dev_info(&dev->dev, "AER: Device recovery failed\n");
547 * handle_error_source - handle logging error into an event log
548 * @aerdev: pointer to pcie_device data structure of the root port
549 * @dev: pointer to pci_dev data structure of error source device
550 * @info: comprehensive error information
552 * Invoked when an error being detected by Root Port.
554 static void handle_error_source(struct pcie_device *aerdev,
556 struct aer_err_info *info)
560 if (info->severity == AER_CORRECTABLE) {
562 * Correctable error does not need software intevention.
563 * No need to go through error recovery process.
565 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
567 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
570 do_recovery(dev, info->severity);
573 #ifdef CONFIG_ACPI_APEI_PCIEAER
574 static void aer_recover_work_func(struct work_struct *work);
576 #define AER_RECOVER_RING_ORDER 4
577 #define AER_RECOVER_RING_SIZE (1 << AER_RECOVER_RING_ORDER)
579 struct aer_recover_entry
587 static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
588 AER_RECOVER_RING_SIZE);
590 * Mutual exclusion for writers of aer_recover_ring, reader side don't
591 * need lock, because there is only one reader and lock is not needed
592 * between reader and writer.
594 static DEFINE_SPINLOCK(aer_recover_ring_lock);
595 static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
597 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
601 struct aer_recover_entry entry = {
605 .severity = severity,
608 spin_lock_irqsave(&aer_recover_ring_lock, flags);
609 if (kfifo_put(&aer_recover_ring, &entry))
610 schedule_work(&aer_recover_work);
612 pr_err("AER recover: Buffer overflow when recovering AER for %04x:%02x:%02x:%x\n",
613 domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
614 spin_unlock_irqrestore(&aer_recover_ring_lock, flags);
616 EXPORT_SYMBOL_GPL(aer_recover_queue);
618 static void aer_recover_work_func(struct work_struct *work)
620 struct aer_recover_entry entry;
621 struct pci_dev *pdev;
623 while (kfifo_get(&aer_recover_ring, &entry)) {
624 pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
627 pr_err("AER recover: Can not find pci_dev for %04x:%02x:%02x:%x\n",
628 entry.domain, entry.bus,
629 PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
632 do_recovery(pdev, entry.severity);
638 * get_device_error_info - read error status from dev and store it to info
639 * @dev: pointer to the device expected to have a error record
640 * @info: pointer to structure to store the error record
642 * Return 1 on success, 0 on error.
644 * Note that @info is reused among all error devices. Clear fields properly.
646 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
650 /* Must reset in this function */
652 info->tlp_header_valid = 0;
654 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
656 /* The device might not support AER */
660 if (info->severity == AER_CORRECTABLE) {
661 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
663 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
665 if (!(info->status & ~info->mask))
667 } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
668 info->severity == AER_NONFATAL) {
670 /* Link is still healthy for IO reads */
671 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
673 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
675 if (!(info->status & ~info->mask))
678 /* Get First Error Pointer */
679 pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
680 info->first_error = PCI_ERR_CAP_FEP(temp);
682 if (info->status & AER_LOG_TLP_MASKS) {
683 info->tlp_header_valid = 1;
684 pci_read_config_dword(dev,
685 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
686 pci_read_config_dword(dev,
687 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
688 pci_read_config_dword(dev,
689 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
690 pci_read_config_dword(dev,
691 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
698 static inline void aer_process_err_devices(struct pcie_device *p_device,
699 struct aer_err_info *e_info)
703 /* Report all before handle them, not to lost records by reset etc. */
704 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
705 if (get_device_error_info(e_info->dev[i], e_info))
706 aer_print_error(e_info->dev[i], e_info);
708 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
709 if (get_device_error_info(e_info->dev[i], e_info))
710 handle_error_source(p_device, e_info->dev[i], e_info);
715 * aer_isr_one_error - consume an error detected by root port
716 * @p_device: pointer to error root port service device
717 * @e_src: pointer to an error source
719 static void aer_isr_one_error(struct pcie_device *p_device,
720 struct aer_err_source *e_src)
722 struct aer_err_info *e_info;
724 /* struct aer_err_info might be big, so we allocate it with slab */
725 e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL);
727 dev_printk(KERN_DEBUG, &p_device->port->dev,
728 "Can't allocate mem when processing AER errors\n");
733 * There is a possibility that both correctable error and
734 * uncorrectable error being logged. Report correctable error first.
736 if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
737 e_info->id = ERR_COR_ID(e_src->id);
738 e_info->severity = AER_CORRECTABLE;
740 if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
741 e_info->multi_error_valid = 1;
743 e_info->multi_error_valid = 0;
745 aer_print_port_info(p_device->port, e_info);
747 if (find_source_device(p_device->port, e_info))
748 aer_process_err_devices(p_device, e_info);
751 if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
752 e_info->id = ERR_UNCOR_ID(e_src->id);
754 if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
755 e_info->severity = AER_FATAL;
757 e_info->severity = AER_NONFATAL;
759 if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
760 e_info->multi_error_valid = 1;
762 e_info->multi_error_valid = 0;
764 aer_print_port_info(p_device->port, e_info);
766 if (find_source_device(p_device->port, e_info))
767 aer_process_err_devices(p_device, e_info);
774 * get_e_source - retrieve an error source
775 * @rpc: pointer to the root port which holds an error
776 * @e_src: pointer to store retrieved error source
778 * Return 1 if an error source is retrieved, otherwise 0.
780 * Invoked by DPC handler to consume an error.
782 static int get_e_source(struct aer_rpc *rpc, struct aer_err_source *e_src)
786 /* Lock access to Root error producer/consumer index */
787 spin_lock_irqsave(&rpc->e_lock, flags);
788 if (rpc->prod_idx == rpc->cons_idx) {
789 spin_unlock_irqrestore(&rpc->e_lock, flags);
793 *e_src = rpc->e_sources[rpc->cons_idx];
795 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
797 spin_unlock_irqrestore(&rpc->e_lock, flags);
803 * aer_isr - consume errors detected by root port
804 * @work: definition of this work item
806 * Invoked, as DPC, when root port records new detected error
808 void aer_isr(struct work_struct *work)
810 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
811 struct pcie_device *p_device = rpc->rpd;
812 struct aer_err_source uninitialized_var(e_src);
814 mutex_lock(&rpc->rpc_mutex);
815 while (get_e_source(rpc, &e_src))
816 aer_isr_one_error(p_device, &e_src);
817 mutex_unlock(&rpc->rpc_mutex);
819 wake_up(&rpc->wait_release);
823 * aer_init - provide AER initialization
824 * @dev: pointer to AER pcie device
826 * Invoked when AER service driver is loaded.
828 int aer_init(struct pcie_device *dev)
831 dev_printk(KERN_DEBUG, &dev->device,
832 "aerdrv forceload requested.\n");
833 pcie_aer_force_firmware_first(dev->port, 0);