• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * drivers/pci/pcie/aer/aerdrv_core.c
3  *
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
6  * for more details.
7  *
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.
12  *
13  * Copyright (C) 2006 Intel Corp.
14  *	Tom Long Nguyen (tom.l.nguyen@intel.com)
15  *	Zhang Yanmin (yanmin.zhang@intel.com)
16  *
17  */
18 
19 #include <linux/module.h>
20 #include <linux/pci.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/pm.h>
24 #include <linux/suspend.h>
25 #include <linux/delay.h>
26 #include <linux/slab.h>
27 #include <linux/kfifo.h>
28 #include "aerdrv.h"
29 
30 #define	PCI_EXP_AER_FLAGS	(PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \
31 				 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE)
32 
pci_enable_pcie_error_reporting(struct pci_dev * dev)33 int pci_enable_pcie_error_reporting(struct pci_dev *dev)
34 {
35 	if (pcie_aer_get_firmware_first(dev))
36 		return -EIO;
37 
38 	if (!dev->aer_cap)
39 		return -EIO;
40 
41 	return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
42 }
43 EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
44 
pci_disable_pcie_error_reporting(struct pci_dev * dev)45 int pci_disable_pcie_error_reporting(struct pci_dev *dev)
46 {
47 	if (pcie_aer_get_firmware_first(dev))
48 		return -EIO;
49 
50 	return pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
51 					  PCI_EXP_AER_FLAGS);
52 }
53 EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
54 
pci_cleanup_aer_uncorrect_error_status(struct pci_dev * dev)55 int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
56 {
57 	int pos;
58 	u32 status;
59 
60 	pos = dev->aer_cap;
61 	if (!pos)
62 		return -EIO;
63 
64 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
65 	if (status)
66 		pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
67 
68 	return 0;
69 }
70 EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
71 
pci_cleanup_aer_error_status_regs(struct pci_dev * dev)72 int pci_cleanup_aer_error_status_regs(struct pci_dev *dev)
73 {
74 	int pos;
75 	u32 status;
76 	int port_type;
77 
78 	if (!pci_is_pcie(dev))
79 		return -ENODEV;
80 
81 	pos = dev->aer_cap;
82 	if (!pos)
83 		return -EIO;
84 
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);
89 	}
90 
91 	pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
92 	pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
93 
94 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
95 	pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
96 
97 	return 0;
98 }
99 
pci_aer_init(struct pci_dev * dev)100 int pci_aer_init(struct pci_dev *dev)
101 {
102 	dev->aer_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
103 	return pci_cleanup_aer_error_status_regs(dev);
104 }
105 
106 /**
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
110  */
add_error_device(struct aer_err_info * e_info,struct pci_dev * dev)111 static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
112 {
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++;
116 		return 0;
117 	}
118 	return -ENOSPC;
119 }
120 
121 /**
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
125  */
is_error_source(struct pci_dev * dev,struct aer_err_info * e_info)126 static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info)
127 {
128 	int pos;
129 	u32 status, mask;
130 	u16 reg16;
131 
132 	/*
133 	 * When bus id is equal to 0, it might be a bad id
134 	 * reported by root port.
135 	 */
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))
140 			return true;
141 
142 		/* Continue id comparing if there is no multiple error */
143 		if (!e_info->multi_error_valid)
144 			return false;
145 	}
146 
147 	/*
148 	 * When either
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.
154 	 */
155 	if (atomic_read(&dev->enable_cnt) == 0)
156 		return false;
157 
158 	/* Check if AER is enabled */
159 	pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &reg16);
160 	if (!(reg16 & PCI_EXP_AER_FLAGS))
161 		return false;
162 
163 	pos = dev->aer_cap;
164 	if (!pos)
165 		return false;
166 
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);
171 	} else {
172 		pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
173 		pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
174 	}
175 	if (status & ~mask)
176 		return true;
177 
178 	return false;
179 }
180 
find_device_iter(struct pci_dev * dev,void * data)181 static int find_device_iter(struct pci_dev *dev, void *data)
182 {
183 	struct aer_err_info *e_info = (struct aer_err_info *)data;
184 
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? */
190 			return 1;
191 		}
192 
193 		/* If there is only a single error, stop iteration */
194 		if (!e_info->multi_error_valid)
195 			return 1;
196 	}
197 	return 0;
198 }
199 
200 /**
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
204  *
205  * Return true if found.
206  *
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.
211  */
find_source_device(struct pci_dev * parent,struct aer_err_info * e_info)212 static bool find_source_device(struct pci_dev *parent,
213 		struct aer_err_info *e_info)
214 {
215 	struct pci_dev *dev = parent;
216 	int result;
217 
218 	/* Must reset in this function */
219 	e_info->error_dev_num = 0;
220 
221 	/* Is Root Port an agent that sends error message? */
222 	result = find_device_iter(dev, e_info);
223 	if (result)
224 		return true;
225 
226 	pci_walk_bus(parent->subordinate, find_device_iter, e_info);
227 
228 	if (!e_info->error_dev_num) {
229 		dev_printk(KERN_DEBUG, &parent->dev,
230 				"can't find device of ID%04x\n",
231 				e_info->id);
232 		return false;
233 	}
234 	return true;
235 }
236 
report_error_detected(struct pci_dev * dev,void * data)237 static int report_error_detected(struct pci_dev *dev, void *data)
238 {
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;
243 
244 	device_lock(&dev->dev);
245 	dev->error_state = result_data->state;
246 
247 	if (!dev->driver ||
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) {
252 			/*
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
257 			 * its hw state.
258 			 */
259 			dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
260 				   dev->driver ?
261 				   "no AER-aware driver" : "no driver");
262 		}
263 
264 		/*
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
271 		 * without recovery.
272 		 */
273 
274 		if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE)
275 			vote = PCI_ERS_RESULT_NO_AER_DRIVER;
276 		else
277 			vote = PCI_ERS_RESULT_NONE;
278 	} else {
279 		err_handler = dev->driver->err_handler;
280 		vote = err_handler->error_detected(dev, result_data->state);
281 	}
282 
283 	result_data->result = merge_result(result_data->result, vote);
284 	device_unlock(&dev->dev);
285 	return 0;
286 }
287 
report_mmio_enabled(struct pci_dev * dev,void * data)288 static int report_mmio_enabled(struct pci_dev *dev, void *data)
289 {
290 	pci_ers_result_t vote;
291 	const struct pci_error_handlers *err_handler;
292 	struct aer_broadcast_data *result_data;
293 	result_data = (struct aer_broadcast_data *) data;
294 
295 	device_lock(&dev->dev);
296 	if (!dev->driver ||
297 		!dev->driver->err_handler ||
298 		!dev->driver->err_handler->mmio_enabled)
299 		goto out;
300 
301 	err_handler = dev->driver->err_handler;
302 	vote = err_handler->mmio_enabled(dev);
303 	result_data->result = merge_result(result_data->result, vote);
304 out:
305 	device_unlock(&dev->dev);
306 	return 0;
307 }
308 
report_slot_reset(struct pci_dev * dev,void * data)309 static int report_slot_reset(struct pci_dev *dev, void *data)
310 {
311 	pci_ers_result_t vote;
312 	const struct pci_error_handlers *err_handler;
313 	struct aer_broadcast_data *result_data;
314 	result_data = (struct aer_broadcast_data *) data;
315 
316 	device_lock(&dev->dev);
317 	if (!dev->driver ||
318 		!dev->driver->err_handler ||
319 		!dev->driver->err_handler->slot_reset)
320 		goto out;
321 
322 	err_handler = dev->driver->err_handler;
323 	vote = err_handler->slot_reset(dev);
324 	result_data->result = merge_result(result_data->result, vote);
325 out:
326 	device_unlock(&dev->dev);
327 	return 0;
328 }
329 
report_resume(struct pci_dev * dev,void * data)330 static int report_resume(struct pci_dev *dev, void *data)
331 {
332 	const struct pci_error_handlers *err_handler;
333 
334 	device_lock(&dev->dev);
335 	dev->error_state = pci_channel_io_normal;
336 
337 	if (!dev->driver ||
338 		!dev->driver->err_handler ||
339 		!dev->driver->err_handler->resume)
340 		goto out;
341 
342 	err_handler = dev->driver->err_handler;
343 	err_handler->resume(dev);
344 out:
345 	device_unlock(&dev->dev);
346 	return 0;
347 }
348 
349 /**
350  * broadcast_error_message - handle message broadcast to downstream drivers
351  * @dev: pointer to from where in a hierarchy message is broadcasted down
352  * @state: error state
353  * @error_mesg: message to print
354  * @cb: callback to be broadcasted
355  *
356  * Invoked during error recovery process. Once being invoked, the content
357  * of error severity will be broadcasted to all downstream drivers in a
358  * hierarchy in question.
359  */
broadcast_error_message(struct pci_dev * dev,enum pci_channel_state state,char * error_mesg,int (* cb)(struct pci_dev *,void *))360 static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
361 	enum pci_channel_state state,
362 	char *error_mesg,
363 	int (*cb)(struct pci_dev *, void *))
364 {
365 	struct aer_broadcast_data result_data;
366 
367 	dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
368 	result_data.state = state;
369 	if (cb == report_error_detected)
370 		result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
371 	else
372 		result_data.result = PCI_ERS_RESULT_RECOVERED;
373 
374 	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
375 		/*
376 		 * If the error is reported by a bridge, we think this error
377 		 * is related to the downstream link of the bridge, so we
378 		 * do error recovery on all subordinates of the bridge instead
379 		 * of the bridge and clear the error status of the bridge.
380 		 */
381 		if (cb == report_error_detected)
382 			dev->error_state = state;
383 		pci_walk_bus(dev->subordinate, cb, &result_data);
384 		if (cb == report_resume) {
385 			pci_cleanup_aer_uncorrect_error_status(dev);
386 			dev->error_state = pci_channel_io_normal;
387 		}
388 	} else {
389 		/*
390 		 * If the error is reported by an end point, we think this
391 		 * error is related to the upstream link of the end point.
392 		 */
393 		if (state == pci_channel_io_normal)
394 			/*
395 			 * the error is non fatal so the bus is ok, just invoke
396 			 * the callback for the function that logged the error.
397 			 */
398 			cb(dev, &result_data);
399 		else
400 			pci_walk_bus(dev->bus, cb, &result_data);
401 	}
402 
403 	return result_data.result;
404 }
405 
406 /**
407  * default_reset_link - default reset function
408  * @dev: pointer to pci_dev data structure
409  *
410  * Invoked when performing link reset on a Downstream Port or a
411  * Root Port with no aer driver.
412  */
default_reset_link(struct pci_dev * dev)413 static pci_ers_result_t default_reset_link(struct pci_dev *dev)
414 {
415 	pci_reset_bridge_secondary_bus(dev);
416 	dev_printk(KERN_DEBUG, &dev->dev, "downstream link has been reset\n");
417 	return PCI_ERS_RESULT_RECOVERED;
418 }
419 
find_aer_service_iter(struct device * device,void * data)420 static int find_aer_service_iter(struct device *device, void *data)
421 {
422 	struct pcie_port_service_driver *service_driver, **drv;
423 
424 	drv = (struct pcie_port_service_driver **) data;
425 
426 	if (device->bus == &pcie_port_bus_type && device->driver) {
427 		service_driver = to_service_driver(device->driver);
428 		if (service_driver->service == PCIE_PORT_SERVICE_AER) {
429 			*drv = service_driver;
430 			return 1;
431 		}
432 	}
433 
434 	return 0;
435 }
436 
find_aer_service(struct pci_dev * dev)437 static struct pcie_port_service_driver *find_aer_service(struct pci_dev *dev)
438 {
439 	struct pcie_port_service_driver *drv = NULL;
440 
441 	device_for_each_child(&dev->dev, &drv, find_aer_service_iter);
442 
443 	return drv;
444 }
445 
reset_link(struct pci_dev * dev)446 static pci_ers_result_t reset_link(struct pci_dev *dev)
447 {
448 	struct pci_dev *udev;
449 	pci_ers_result_t status;
450 	struct pcie_port_service_driver *driver;
451 
452 	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
453 		/* Reset this port for all subordinates */
454 		udev = dev;
455 	} else {
456 		/* Reset the upstream component (likely downstream port) */
457 		udev = dev->bus->self;
458 	}
459 
460 	/* Use the aer driver of the component firstly */
461 	driver = find_aer_service(udev);
462 
463 	if (driver && driver->reset_link) {
464 		status = driver->reset_link(udev);
465 	} else if (udev->has_secondary_link) {
466 		status = default_reset_link(udev);
467 	} else {
468 		dev_printk(KERN_DEBUG, &dev->dev,
469 			"no link-reset support at upstream device %s\n",
470 			pci_name(udev));
471 		return PCI_ERS_RESULT_DISCONNECT;
472 	}
473 
474 	if (status != PCI_ERS_RESULT_RECOVERED) {
475 		dev_printk(KERN_DEBUG, &dev->dev,
476 			"link reset at upstream device %s failed\n",
477 			pci_name(udev));
478 		return PCI_ERS_RESULT_DISCONNECT;
479 	}
480 
481 	return status;
482 }
483 
484 /**
485  * do_recovery - handle nonfatal/fatal error recovery process
486  * @dev: pointer to a pci_dev data structure of agent detecting an error
487  * @severity: error severity type
488  *
489  * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
490  * error detected message to all downstream drivers within a hierarchy in
491  * question and return the returned code.
492  */
do_recovery(struct pci_dev * dev,int severity)493 static void do_recovery(struct pci_dev *dev, int severity)
494 {
495 	pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
496 	enum pci_channel_state state;
497 
498 	if (severity == AER_FATAL)
499 		state = pci_channel_io_frozen;
500 	else
501 		state = pci_channel_io_normal;
502 
503 	status = broadcast_error_message(dev,
504 			state,
505 			"error_detected",
506 			report_error_detected);
507 
508 	if (severity == AER_FATAL) {
509 		result = reset_link(dev);
510 		if (result != PCI_ERS_RESULT_RECOVERED)
511 			goto failed;
512 	}
513 
514 	if (status == PCI_ERS_RESULT_CAN_RECOVER)
515 		status = broadcast_error_message(dev,
516 				state,
517 				"mmio_enabled",
518 				report_mmio_enabled);
519 
520 	if (status == PCI_ERS_RESULT_NEED_RESET) {
521 		/*
522 		 * TODO: Should call platform-specific
523 		 * functions to reset slot before calling
524 		 * drivers' slot_reset callbacks?
525 		 */
526 		status = broadcast_error_message(dev,
527 				state,
528 				"slot_reset",
529 				report_slot_reset);
530 	}
531 
532 	if (status != PCI_ERS_RESULT_RECOVERED)
533 		goto failed;
534 
535 	broadcast_error_message(dev,
536 				state,
537 				"resume",
538 				report_resume);
539 
540 	dev_info(&dev->dev, "AER: Device recovery successful\n");
541 	return;
542 
543 failed:
544 	/* TODO: Should kernel panic here? */
545 	dev_info(&dev->dev, "AER: Device recovery failed\n");
546 }
547 
548 /**
549  * handle_error_source - handle logging error into an event log
550  * @aerdev: pointer to pcie_device data structure of the root port
551  * @dev: pointer to pci_dev data structure of error source device
552  * @info: comprehensive error information
553  *
554  * Invoked when an error being detected by Root Port.
555  */
handle_error_source(struct pcie_device * aerdev,struct pci_dev * dev,struct aer_err_info * info)556 static void handle_error_source(struct pcie_device *aerdev,
557 	struct pci_dev *dev,
558 	struct aer_err_info *info)
559 {
560 	int pos;
561 
562 	if (info->severity == AER_CORRECTABLE) {
563 		/*
564 		 * Correctable error does not need software intervention.
565 		 * No need to go through error recovery process.
566 		 */
567 		pos = dev->aer_cap;
568 		if (pos)
569 			pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
570 					info->status);
571 	} else
572 		do_recovery(dev, info->severity);
573 }
574 
575 #ifdef CONFIG_ACPI_APEI_PCIEAER
576 static void aer_recover_work_func(struct work_struct *work);
577 
578 #define AER_RECOVER_RING_ORDER		4
579 #define AER_RECOVER_RING_SIZE		(1 << AER_RECOVER_RING_ORDER)
580 
581 struct aer_recover_entry {
582 	u8	bus;
583 	u8	devfn;
584 	u16	domain;
585 	int	severity;
586 	struct aer_capability_regs *regs;
587 };
588 
589 static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
590 		    AER_RECOVER_RING_SIZE);
591 /*
592  * Mutual exclusion for writers of aer_recover_ring, reader side don't
593  * need lock, because there is only one reader and lock is not needed
594  * between reader and writer.
595  */
596 static DEFINE_SPINLOCK(aer_recover_ring_lock);
597 static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
598 
aer_recover_queue(int domain,unsigned int bus,unsigned int devfn,int severity,struct aer_capability_regs * aer_regs)599 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
600 		       int severity, struct aer_capability_regs *aer_regs)
601 {
602 	unsigned long flags;
603 	struct aer_recover_entry entry = {
604 		.bus		= bus,
605 		.devfn		= devfn,
606 		.domain		= domain,
607 		.severity	= severity,
608 		.regs		= aer_regs,
609 	};
610 
611 	spin_lock_irqsave(&aer_recover_ring_lock, flags);
612 	if (kfifo_put(&aer_recover_ring, entry))
613 		schedule_work(&aer_recover_work);
614 	else
615 		pr_err("AER recover: Buffer overflow when recovering AER for %04x:%02x:%02x:%x\n",
616 		       domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
617 	spin_unlock_irqrestore(&aer_recover_ring_lock, flags);
618 }
619 EXPORT_SYMBOL_GPL(aer_recover_queue);
620 
aer_recover_work_func(struct work_struct * work)621 static void aer_recover_work_func(struct work_struct *work)
622 {
623 	struct aer_recover_entry entry;
624 	struct pci_dev *pdev;
625 
626 	while (kfifo_get(&aer_recover_ring, &entry)) {
627 		pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
628 						   entry.devfn);
629 		if (!pdev) {
630 			pr_err("AER recover: Can not find pci_dev for %04x:%02x:%02x:%x\n",
631 			       entry.domain, entry.bus,
632 			       PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
633 			continue;
634 		}
635 		cper_print_aer(pdev, entry.severity, entry.regs);
636 		do_recovery(pdev, entry.severity);
637 		pci_dev_put(pdev);
638 	}
639 }
640 #endif
641 
642 /**
643  * get_device_error_info - read error status from dev and store it to info
644  * @dev: pointer to the device expected to have a error record
645  * @info: pointer to structure to store the error record
646  *
647  * Return 1 on success, 0 on error.
648  *
649  * Note that @info is reused among all error devices. Clear fields properly.
650  */
get_device_error_info(struct pci_dev * dev,struct aer_err_info * info)651 static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
652 {
653 	int pos, temp;
654 
655 	/* Must reset in this function */
656 	info->status = 0;
657 	info->tlp_header_valid = 0;
658 
659 	pos = dev->aer_cap;
660 
661 	/* The device might not support AER */
662 	if (!pos)
663 		return 1;
664 
665 	if (info->severity == AER_CORRECTABLE) {
666 		pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
667 			&info->status);
668 		pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
669 			&info->mask);
670 		if (!(info->status & ~info->mask))
671 			return 0;
672 	} else if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
673 		info->severity == AER_NONFATAL) {
674 
675 		/* Link is still healthy for IO reads */
676 		pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
677 			&info->status);
678 		pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
679 			&info->mask);
680 		if (!(info->status & ~info->mask))
681 			return 0;
682 
683 		/* Get First Error Pointer */
684 		pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
685 		info->first_error = PCI_ERR_CAP_FEP(temp);
686 
687 		if (info->status & AER_LOG_TLP_MASKS) {
688 			info->tlp_header_valid = 1;
689 			pci_read_config_dword(dev,
690 				pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
691 			pci_read_config_dword(dev,
692 				pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
693 			pci_read_config_dword(dev,
694 				pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
695 			pci_read_config_dword(dev,
696 				pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
697 		}
698 	}
699 
700 	return 1;
701 }
702 
aer_process_err_devices(struct pcie_device * p_device,struct aer_err_info * e_info)703 static inline void aer_process_err_devices(struct pcie_device *p_device,
704 			struct aer_err_info *e_info)
705 {
706 	int i;
707 
708 	/* Report all before handle them, not to lost records by reset etc. */
709 	for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
710 		if (get_device_error_info(e_info->dev[i], e_info))
711 			aer_print_error(e_info->dev[i], e_info);
712 	}
713 	for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
714 		if (get_device_error_info(e_info->dev[i], e_info))
715 			handle_error_source(p_device, e_info->dev[i], e_info);
716 	}
717 }
718 
719 /**
720  * aer_isr_one_error - consume an error detected by root port
721  * @p_device: pointer to error root port service device
722  * @e_src: pointer to an error source
723  */
aer_isr_one_error(struct pcie_device * p_device,struct aer_err_source * e_src)724 static void aer_isr_one_error(struct pcie_device *p_device,
725 		struct aer_err_source *e_src)
726 {
727 	struct aer_rpc *rpc = get_service_data(p_device);
728 	struct aer_err_info *e_info = &rpc->e_info;
729 
730 	/*
731 	 * There is a possibility that both correctable error and
732 	 * uncorrectable error being logged. Report correctable error first.
733 	 */
734 	if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
735 		e_info->id = ERR_COR_ID(e_src->id);
736 		e_info->severity = AER_CORRECTABLE;
737 
738 		if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
739 			e_info->multi_error_valid = 1;
740 		else
741 			e_info->multi_error_valid = 0;
742 
743 		aer_print_port_info(p_device->port, e_info);
744 
745 		if (find_source_device(p_device->port, e_info))
746 			aer_process_err_devices(p_device, e_info);
747 	}
748 
749 	if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
750 		e_info->id = ERR_UNCOR_ID(e_src->id);
751 
752 		if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
753 			e_info->severity = AER_FATAL;
754 		else
755 			e_info->severity = AER_NONFATAL;
756 
757 		if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
758 			e_info->multi_error_valid = 1;
759 		else
760 			e_info->multi_error_valid = 0;
761 
762 		aer_print_port_info(p_device->port, e_info);
763 
764 		if (find_source_device(p_device->port, e_info))
765 			aer_process_err_devices(p_device, e_info);
766 	}
767 }
768 
769 /**
770  * get_e_source - retrieve an error source
771  * @rpc: pointer to the root port which holds an error
772  * @e_src: pointer to store retrieved error source
773  *
774  * Return 1 if an error source is retrieved, otherwise 0.
775  *
776  * Invoked by DPC handler to consume an error.
777  */
get_e_source(struct aer_rpc * rpc,struct aer_err_source * e_src)778 static int get_e_source(struct aer_rpc *rpc, struct aer_err_source *e_src)
779 {
780 	unsigned long flags;
781 
782 	/* Lock access to Root error producer/consumer index */
783 	spin_lock_irqsave(&rpc->e_lock, flags);
784 	if (rpc->prod_idx == rpc->cons_idx) {
785 		spin_unlock_irqrestore(&rpc->e_lock, flags);
786 		return 0;
787 	}
788 
789 	*e_src = rpc->e_sources[rpc->cons_idx];
790 	rpc->cons_idx++;
791 	if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
792 		rpc->cons_idx = 0;
793 	spin_unlock_irqrestore(&rpc->e_lock, flags);
794 
795 	return 1;
796 }
797 
798 /**
799  * aer_isr - consume errors detected by root port
800  * @work: definition of this work item
801  *
802  * Invoked, as DPC, when root port records new detected error
803  */
aer_isr(struct work_struct * work)804 void aer_isr(struct work_struct *work)
805 {
806 	struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
807 	struct pcie_device *p_device = rpc->rpd;
808 	struct aer_err_source uninitialized_var(e_src);
809 
810 	mutex_lock(&rpc->rpc_mutex);
811 	while (get_e_source(rpc, &e_src))
812 		aer_isr_one_error(p_device, &e_src);
813 	mutex_unlock(&rpc->rpc_mutex);
814 }
815