blob: d3f6df40b0d33bf3801824bf1718350c881e92d0 [file] [log] [blame]
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +08001/*
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>
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080025#include <linux/delay.h>
26#include "aerdrv.h"
27
28static int forceload;
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +080029static int nosourceid;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080030module_param(forceload, bool, 0);
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +080031module_param(nosourceid, bool, 0);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080032
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080033int pci_enable_pcie_error_reporting(struct pci_dev *dev)
34{
35 u16 reg16 = 0;
36 int pos;
37
Yu Zhao270c66b2008-10-19 20:35:20 +080038 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080039 if (!pos)
40 return -EIO;
41
Yu Zhao270c66b2008-10-19 20:35:20 +080042 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
Jesse Barnes09276782008-10-18 17:33:19 -070043 if (!pos)
44 return -EIO;
45
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080046 pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
47 reg16 = reg16 |
48 PCI_EXP_DEVCTL_CERE |
49 PCI_EXP_DEVCTL_NFERE |
50 PCI_EXP_DEVCTL_FERE |
51 PCI_EXP_DEVCTL_URRE;
Hidetoshi Setoc9a91882009-09-07 17:07:29 +090052 pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
53
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080054 return 0;
55}
Hidetoshi Setoc9a91882009-09-07 17:07:29 +090056EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080057
58int pci_disable_pcie_error_reporting(struct pci_dev *dev)
59{
60 u16 reg16 = 0;
61 int pos;
62
63 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
64 if (!pos)
65 return -EIO;
66
67 pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
68 reg16 = reg16 & ~(PCI_EXP_DEVCTL_CERE |
69 PCI_EXP_DEVCTL_NFERE |
70 PCI_EXP_DEVCTL_FERE |
71 PCI_EXP_DEVCTL_URRE);
Hidetoshi Setoc9a91882009-09-07 17:07:29 +090072 pci_write_config_word(dev, pos+PCI_EXP_DEVCTL, reg16);
73
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080074 return 0;
75}
Hidetoshi Setoc9a91882009-09-07 17:07:29 +090076EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080077
78int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
79{
80 int pos;
81 u32 status, mask;
82
Jesse Barnes09276782008-10-18 17:33:19 -070083 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080084 if (!pos)
85 return -EIO;
86
87 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
88 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
89 if (dev->error_state == pci_channel_io_normal)
90 status &= ~mask; /* Clear corresponding nonfatal bits */
91 else
92 status &= mask; /* Clear corresponding fatal bits */
93 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
94
95 return 0;
96}
Hidetoshi Setoc9a91882009-09-07 17:07:29 +090097EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +080098
Adrian Bunk21c68472008-02-04 23:50:11 -080099#if 0
Stephen Hemmingerf0dce412007-06-06 11:50:34 +0800100int pci_cleanup_aer_correct_error_status(struct pci_dev *dev)
101{
102 int pos;
103 u32 status;
104
Jesse Barnes09276782008-10-18 17:33:19 -0700105 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
Stephen Hemmingerf0dce412007-06-06 11:50:34 +0800106 if (!pos)
107 return -EIO;
108
109 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
110 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, status);
111
112 return 0;
113}
Adrian Bunk21c68472008-02-04 23:50:11 -0800114#endif /* 0 */
Stephen Hemmingerf0dce412007-06-06 11:50:34 +0800115
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800116static int set_device_error_reporting(struct pci_dev *dev, void *data)
Andrew Patterson1f9f13c2009-02-20 16:04:59 -0700117{
118 bool enable = *((bool *)data);
119
Andrew Patterson43c16402009-04-22 16:52:09 -0600120 if (dev->pcie_type == PCIE_RC_PORT ||
121 dev->pcie_type == PCIE_SW_UPSTREAM_PORT ||
122 dev->pcie_type == PCIE_SW_DOWNSTREAM_PORT) {
123 if (enable)
124 pci_enable_pcie_error_reporting(dev);
125 else
126 pci_disable_pcie_error_reporting(dev);
127 }
Andrew Patterson1f9f13c2009-02-20 16:04:59 -0700128
129 if (enable)
Andrew Patterson43c16402009-04-22 16:52:09 -0600130 pcie_set_ecrc_checking(dev);
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800131
132 return 0;
Andrew Patterson1f9f13c2009-02-20 16:04:59 -0700133}
134
135/**
136 * set_downstream_devices_error_reporting - enable/disable the error reporting bits on the root port and its downstream ports.
137 * @dev: pointer to root port's pci_dev data structure
138 * @enable: true = enable error reporting, false = disable error reporting.
139 */
140static void set_downstream_devices_error_reporting(struct pci_dev *dev,
141 bool enable)
142{
143 set_device_error_reporting(dev, &enable);
Alex Chiangcb4cb4a2009-03-05 19:28:40 -0700144
145 if (!dev->subordinate)
146 return;
Andrew Patterson1f9f13c2009-02-20 16:04:59 -0700147 pci_walk_bus(dev->subordinate, set_device_error_reporting, &enable);
148}
149
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800150static inline int compare_device_id(struct pci_dev *dev,
151 struct aer_err_info *e_info)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800152{
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800153 if (e_info->id == ((dev->bus->number << 8) | dev->devfn)) {
154 /*
155 * Device ID match
156 */
157 return 1;
158 }
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800159
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800160 return 0;
161}
162
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800163static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
164{
165 if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
166 e_info->dev[e_info->error_dev_num] = dev;
167 e_info->error_dev_num++;
168 return 1;
Hidetoshi Setoc9a91882009-09-07 17:07:29 +0900169 }
170
171 return 0;
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800172}
173
174
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800175#define PCI_BUS(x) (((x) >> 8) & 0xff)
176
177static int find_device_iter(struct pci_dev *dev, void *data)
178{
179 int pos;
180 u32 status;
181 u32 mask;
182 u16 reg16;
183 int result;
184 struct aer_err_info *e_info = (struct aer_err_info *)data;
185
186 /*
187 * When bus id is equal to 0, it might be a bad id
188 * reported by root port.
189 */
190 if (!nosourceid && (PCI_BUS(e_info->id) != 0)) {
191 result = compare_device_id(dev, e_info);
192 if (result)
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800193 add_error_device(e_info, dev);
194
195 /*
196 * If there is no multiple error, we stop
197 * or continue based on the id comparing.
198 */
199 if (!(e_info->flags & AER_MULTI_ERROR_VALID_FLAG))
200 return result;
201
202 /*
203 * If there are multiple errors and id does match,
204 * We need continue to search other devices under
205 * the root port. Return 0 means that.
206 */
207 if (result)
208 return 0;
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800209 }
210
211 /*
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800212 * When either
213 * 1) nosourceid==y;
214 * 2) bus id is equal to 0. Some ports might lose the bus
215 * id of error source id;
216 * 3) There are multiple errors and prior id comparing fails;
217 * We check AER status registers to find the initial reporter.
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800218 */
219 if (atomic_read(&dev->enable_cnt) == 0)
220 return 0;
221 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
222 if (!pos)
223 return 0;
224 /* Check if AER is enabled */
225 pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, &reg16);
226 if (!(reg16 & (
227 PCI_EXP_DEVCTL_CERE |
228 PCI_EXP_DEVCTL_NFERE |
229 PCI_EXP_DEVCTL_FERE |
230 PCI_EXP_DEVCTL_URRE)))
231 return 0;
232 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
233 if (!pos)
234 return 0;
235
236 status = 0;
237 mask = 0;
238 if (e_info->severity == AER_CORRECTABLE) {
239 pci_read_config_dword(dev,
240 pos + PCI_ERR_COR_STATUS,
241 &status);
242 pci_read_config_dword(dev,
243 pos + PCI_ERR_COR_MASK,
244 &mask);
245 if (status & ERR_CORRECTABLE_ERROR_MASK & ~mask) {
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800246 add_error_device(e_info, dev);
247 goto added;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800248 }
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800249 } else {
250 pci_read_config_dword(dev,
251 pos + PCI_ERR_UNCOR_STATUS,
252 &status);
253 pci_read_config_dword(dev,
254 pos + PCI_ERR_UNCOR_MASK,
255 &mask);
256 if (status & ERR_UNCORRECTABLE_ERROR_MASK & ~mask) {
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800257 add_error_device(e_info, dev);
258 goto added;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800259 }
260 }
261
262 return 0;
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800263
264added:
265 if (e_info->flags & AER_MULTI_ERROR_VALID_FLAG)
266 return 0;
267 else
268 return 1;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800269}
270
271/**
272 * find_source_device - search through device hierarchy for source device
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800273 * @parent: pointer to Root Port pci_dev data structure
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800274 * @err_info: including detailed error information such like id
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800275 *
276 * Invoked when error is detected at the Root Port.
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800277 */
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800278static void find_source_device(struct pci_dev *parent,
279 struct aer_err_info *e_info)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800280{
281 struct pci_dev *dev = parent;
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800282 int result;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800283
284 /* Is Root Port an agent that sends error message? */
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800285 result = find_device_iter(dev, e_info);
286 if (result)
287 return;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800288
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800289 pci_walk_bus(parent->subordinate, find_device_iter, e_info);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800290}
291
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800292static int report_error_detected(struct pci_dev *dev, void *data)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800293{
294 pci_ers_result_t vote;
295 struct pci_error_handlers *err_handler;
296 struct aer_broadcast_data *result_data;
297 result_data = (struct aer_broadcast_data *) data;
298
299 dev->error_state = result_data->state;
300
301 if (!dev->driver ||
302 !dev->driver->err_handler ||
303 !dev->driver->err_handler->error_detected) {
304 if (result_data->state == pci_channel_io_frozen &&
305 !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
306 /*
307 * In case of fatal recovery, if one of down-
308 * stream device has no driver. We might be
309 * unable to recover because a later insmod
310 * of a driver for this device is unaware of
311 * its hw state.
312 */
Bjorn Helgaas531f2542008-06-13 10:52:12 -0600313 dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
314 dev->driver ?
315 "no AER-aware driver" : "no driver");
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800316 }
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800317 return 0;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800318 }
319
320 err_handler = dev->driver->err_handler;
321 vote = err_handler->error_detected(dev, result_data->state);
322 result_data->result = merge_result(result_data->result, vote);
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800323 return 0;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800324}
325
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800326static int report_mmio_enabled(struct pci_dev *dev, void *data)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800327{
328 pci_ers_result_t vote;
329 struct pci_error_handlers *err_handler;
330 struct aer_broadcast_data *result_data;
331 result_data = (struct aer_broadcast_data *) data;
332
333 if (!dev->driver ||
334 !dev->driver->err_handler ||
335 !dev->driver->err_handler->mmio_enabled)
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800336 return 0;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800337
338 err_handler = dev->driver->err_handler;
339 vote = err_handler->mmio_enabled(dev);
340 result_data->result = merge_result(result_data->result, vote);
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800341 return 0;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800342}
343
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800344static int report_slot_reset(struct pci_dev *dev, void *data)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800345{
346 pci_ers_result_t vote;
347 struct pci_error_handlers *err_handler;
348 struct aer_broadcast_data *result_data;
349 result_data = (struct aer_broadcast_data *) data;
350
351 if (!dev->driver ||
352 !dev->driver->err_handler ||
353 !dev->driver->err_handler->slot_reset)
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800354 return 0;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800355
356 err_handler = dev->driver->err_handler;
357 vote = err_handler->slot_reset(dev);
358 result_data->result = merge_result(result_data->result, vote);
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800359 return 0;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800360}
361
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800362static int report_resume(struct pci_dev *dev, void *data)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800363{
364 struct pci_error_handlers *err_handler;
365
366 dev->error_state = pci_channel_io_normal;
367
368 if (!dev->driver ||
369 !dev->driver->err_handler ||
Hidetoshi Setob0b801d2008-12-01 16:31:06 +0900370 !dev->driver->err_handler->resume)
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800371 return 0;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800372
373 err_handler = dev->driver->err_handler;
374 err_handler->resume(dev);
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800375 return 0;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800376}
377
378/**
379 * broadcast_error_message - handle message broadcast to downstream drivers
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800380 * @dev: pointer to from where in a hierarchy message is broadcasted down
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800381 * @state: error state
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800382 * @error_mesg: message to print
383 * @cb: callback to be broadcasted
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800384 *
385 * Invoked during error recovery process. Once being invoked, the content
386 * of error severity will be broadcasted to all downstream drivers in a
387 * hierarchy in question.
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800388 */
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800389static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
390 enum pci_channel_state state,
391 char *error_mesg,
Zhang, Yanmin70298c62009-06-16 13:34:38 +0800392 int (*cb)(struct pci_dev *, void *))
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800393{
394 struct aer_broadcast_data result_data;
395
Bjorn Helgaas531f2542008-06-13 10:52:12 -0600396 dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800397 result_data.state = state;
398 if (cb == report_error_detected)
399 result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
400 else
401 result_data.result = PCI_ERS_RESULT_RECOVERED;
402
403 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
404 /*
405 * If the error is reported by a bridge, we think this error
406 * is related to the downstream link of the bridge, so we
407 * do error recovery on all subordinates of the bridge instead
408 * of the bridge and clear the error status of the bridge.
409 */
410 if (cb == report_error_detected)
411 dev->error_state = state;
412 pci_walk_bus(dev->subordinate, cb, &result_data);
413 if (cb == report_resume) {
414 pci_cleanup_aer_uncorrect_error_status(dev);
415 dev->error_state = pci_channel_io_normal;
416 }
Hidetoshi Setoc9a91882009-09-07 17:07:29 +0900417 } else {
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800418 /*
419 * If the error is reported by an end point, we think this
420 * error is related to the upstream link of the end point.
421 */
422 pci_walk_bus(dev->bus, cb, &result_data);
423 }
424
425 return result_data.result;
426}
427
428struct find_aer_service_data {
429 struct pcie_port_service_driver *aer_driver;
430 int is_downstream;
431};
432
433static int find_aer_service_iter(struct device *device, void *data)
434{
435 struct device_driver *driver;
436 struct pcie_port_service_driver *service_driver;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800437 struct find_aer_service_data *result;
438
439 result = (struct find_aer_service_data *) data;
440
441 if (device->bus == &pcie_port_bus_type) {
Rafael J. Wysocki22106362009-01-13 14:46:46 +0100442 struct pcie_port_data *port_data;
443
444 port_data = pci_get_drvdata(to_pcie_device(device)->port);
445 if (port_data->port_type == PCIE_SW_DOWNSTREAM_PORT)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800446 result->is_downstream = 1;
447
448 driver = device->driver;
449 if (driver) {
450 service_driver = to_service_driver(driver);
Rafael J. Wysocki22106362009-01-13 14:46:46 +0100451 if (service_driver->service == PCIE_PORT_SERVICE_AER) {
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800452 result->aer_driver = service_driver;
453 return 1;
454 }
455 }
456 }
457
458 return 0;
459}
460
461static void find_aer_service(struct pci_dev *dev,
462 struct find_aer_service_data *data)
463{
Greg Kroah-Hartmanb19441a2006-08-28 11:43:25 -0700464 int retval;
465 retval = device_for_each_child(&dev->dev, data, find_aer_service_iter);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800466}
467
468static pci_ers_result_t reset_link(struct pcie_device *aerdev,
469 struct pci_dev *dev)
470{
471 struct pci_dev *udev;
472 pci_ers_result_t status;
473 struct find_aer_service_data data;
474
475 if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)
476 udev = dev;
477 else
Hidetoshi Setoc9a91882009-09-07 17:07:29 +0900478 udev = dev->bus->self;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800479
480 data.is_downstream = 0;
481 data.aer_driver = NULL;
482 find_aer_service(udev, &data);
483
484 /*
485 * Use the aer driver of the error agent firstly.
486 * If it hasn't the aer driver, use the root port's
487 */
488 if (!data.aer_driver || !data.aer_driver->reset_link) {
489 if (data.is_downstream &&
490 aerdev->device.driver &&
491 to_service_driver(aerdev->device.driver)->reset_link) {
492 data.aer_driver =
493 to_service_driver(aerdev->device.driver);
494 } else {
Bjorn Helgaas531f2542008-06-13 10:52:12 -0600495 dev_printk(KERN_DEBUG, &dev->dev, "no link-reset "
496 "support\n");
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800497 return PCI_ERS_RESULT_DISCONNECT;
498 }
499 }
500
501 status = data.aer_driver->reset_link(udev);
502 if (status != PCI_ERS_RESULT_RECOVERED) {
Bjorn Helgaas531f2542008-06-13 10:52:12 -0600503 dev_printk(KERN_DEBUG, &dev->dev, "link reset at upstream "
504 "device %s failed\n", pci_name(udev));
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800505 return PCI_ERS_RESULT_DISCONNECT;
506 }
507
508 return status;
509}
510
511/**
512 * do_recovery - handle nonfatal/fatal error recovery process
513 * @aerdev: pointer to a pcie_device data structure of root port
514 * @dev: pointer to a pci_dev data structure of agent detecting an error
515 * @severity: error severity type
516 *
517 * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
518 * error detected message to all downstream drivers within a hierarchy in
519 * question and return the returned code.
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800520 */
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800521static pci_ers_result_t do_recovery(struct pcie_device *aerdev,
522 struct pci_dev *dev,
523 int severity)
524{
525 pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
526 enum pci_channel_state state;
527
528 if (severity == AER_FATAL)
529 state = pci_channel_io_frozen;
530 else
531 state = pci_channel_io_normal;
532
533 status = broadcast_error_message(dev,
534 state,
535 "error_detected",
536 report_error_detected);
537
538 if (severity == AER_FATAL) {
539 result = reset_link(aerdev, dev);
540 if (result != PCI_ERS_RESULT_RECOVERED) {
541 /* TODO: Should panic here? */
542 return result;
543 }
544 }
545
546 if (status == PCI_ERS_RESULT_CAN_RECOVER)
547 status = broadcast_error_message(dev,
548 state,
549 "mmio_enabled",
550 report_mmio_enabled);
551
552 if (status == PCI_ERS_RESULT_NEED_RESET) {
553 /*
554 * TODO: Should call platform-specific
555 * functions to reset slot before calling
556 * drivers' slot_reset callbacks?
557 */
558 status = broadcast_error_message(dev,
559 state,
560 "slot_reset",
561 report_slot_reset);
562 }
563
564 if (status == PCI_ERS_RESULT_RECOVERED)
565 broadcast_error_message(dev,
566 state,
567 "resume",
568 report_resume);
569
570 return status;
571}
572
573/**
574 * handle_error_source - handle logging error into an event log
575 * @aerdev: pointer to pcie_device data structure of the root port
576 * @dev: pointer to pci_dev data structure of error source device
577 * @info: comprehensive error information
578 *
579 * Invoked when an error being detected by Root Port.
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800580 */
Hidetoshi Setoc9a91882009-09-07 17:07:29 +0900581static void handle_error_source(struct pcie_device *aerdev,
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800582 struct pci_dev *dev,
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800583 struct aer_err_info *info)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800584{
585 pci_ers_result_t status = 0;
586 int pos;
587
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800588 if (info->severity == AER_CORRECTABLE) {
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800589 /*
590 * Correctable error does not need software intevention.
591 * No need to go through error recovery process.
592 */
Jesse Barnes09276782008-10-18 17:33:19 -0700593 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800594 if (pos)
595 pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800596 info->status);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800597 } else {
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800598 status = do_recovery(aerdev, dev, info->severity);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800599 if (status == PCI_ERS_RESULT_RECOVERED) {
Bjorn Helgaas531f2542008-06-13 10:52:12 -0600600 dev_printk(KERN_DEBUG, &dev->dev, "AER driver "
601 "successfully recovered\n");
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800602 } else {
603 /* TODO: Should kernel panic here? */
Bjorn Helgaas531f2542008-06-13 10:52:12 -0600604 dev_printk(KERN_DEBUG, &dev->dev, "AER driver didn't "
605 "recover\n");
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800606 }
607 }
608}
609
610/**
611 * aer_enable_rootport - enable Root Port's interrupts when receiving messages
612 * @rpc: pointer to a Root Port data structure
613 *
614 * Invoked when PCIE bus loads AER service driver.
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800615 */
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800616void aer_enable_rootport(struct aer_rpc *rpc)
617{
618 struct pci_dev *pdev = rpc->rpd->port;
619 int pos, aer_pos;
620 u16 reg16;
621 u32 reg32;
622
623 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
624 /* Clear PCIE Capability's Device Status */
625 pci_read_config_word(pdev, pos+PCI_EXP_DEVSTA, &reg16);
626 pci_write_config_word(pdev, pos+PCI_EXP_DEVSTA, reg16);
627
628 /* Disable system error generation in response to error messages */
629 pci_read_config_word(pdev, pos + PCI_EXP_RTCTL, &reg16);
630 reg16 &= ~(SYSTEM_ERROR_INTR_ON_MESG_MASK);
631 pci_write_config_word(pdev, pos + PCI_EXP_RTCTL, reg16);
632
Jesse Barnes09276782008-10-18 17:33:19 -0700633 aer_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800634 /* Clear error status */
635 pci_read_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, &reg32);
636 pci_write_config_dword(pdev, aer_pos + PCI_ERR_ROOT_STATUS, reg32);
637 pci_read_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, &reg32);
638 pci_write_config_dword(pdev, aer_pos + PCI_ERR_COR_STATUS, reg32);
639 pci_read_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, &reg32);
640 pci_write_config_dword(pdev, aer_pos + PCI_ERR_UNCOR_STATUS, reg32);
641
Andrew Patterson1f9f13c2009-02-20 16:04:59 -0700642 /*
643 * Enable error reporting for the root port device and downstream port
644 * devices.
645 */
646 set_downstream_devices_error_reporting(pdev, true);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800647
648 /* Enable Root Port's interrupt in response to error messages */
649 pci_write_config_dword(pdev,
650 aer_pos + PCI_ERR_ROOT_COMMAND,
651 ROOT_PORT_INTR_ON_MESG_MASK);
652}
653
654/**
655 * disable_root_aer - disable Root Port's interrupts when receiving messages
656 * @rpc: pointer to a Root Port data structure
657 *
658 * Invoked when PCIE bus unloads AER service driver.
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800659 */
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800660static void disable_root_aer(struct aer_rpc *rpc)
661{
662 struct pci_dev *pdev = rpc->rpd->port;
663 u32 reg32;
664 int pos;
665
Andrew Patterson1f9f13c2009-02-20 16:04:59 -0700666 /*
667 * Disable error reporting for the root port device and downstream port
668 * devices.
669 */
670 set_downstream_devices_error_reporting(pdev, false);
671
Jesse Barnes09276782008-10-18 17:33:19 -0700672 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ERR);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800673 /* Disable Root's interrupt in response to error messages */
674 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_COMMAND, 0);
675
676 /* Clear Root's error status reg */
677 pci_read_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, &reg32);
678 pci_write_config_dword(pdev, pos + PCI_ERR_ROOT_STATUS, reg32);
679}
680
681/**
682 * get_e_source - retrieve an error source
683 * @rpc: pointer to the root port which holds an error
684 *
685 * Invoked by DPC handler to consume an error.
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800686 */
Hidetoshi Setoc9a91882009-09-07 17:07:29 +0900687static struct aer_err_source *get_e_source(struct aer_rpc *rpc)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800688{
689 struct aer_err_source *e_source;
690 unsigned long flags;
691
692 /* Lock access to Root error producer/consumer index */
693 spin_lock_irqsave(&rpc->e_lock, flags);
694 if (rpc->prod_idx == rpc->cons_idx) {
695 spin_unlock_irqrestore(&rpc->e_lock, flags);
696 return NULL;
697 }
698 e_source = &rpc->e_sources[rpc->cons_idx];
699 rpc->cons_idx++;
700 if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
701 rpc->cons_idx = 0;
702 spin_unlock_irqrestore(&rpc->e_lock, flags);
703
704 return e_source;
705}
706
707static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
708{
709 int pos;
710
Jesse Barnes09276782008-10-18 17:33:19 -0700711 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800712
713 /* The device might not support AER */
714 if (!pos)
715 return AER_SUCCESS;
716
717 if (info->severity == AER_CORRECTABLE) {
718 pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
719 &info->status);
720 if (!(info->status & ERR_CORRECTABLE_ERROR_MASK))
721 return AER_UNSUCCESS;
722 } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
723 info->severity == AER_NONFATAL) {
724
725 /* Link is still healthy for IO reads */
726 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
727 &info->status);
728 if (!(info->status & ERR_UNCORRECTABLE_ERROR_MASK))
729 return AER_UNSUCCESS;
730
731 if (info->status & AER_LOG_TLP_MASKS) {
732 info->flags |= AER_TLP_HEADER_VALID_FLAG;
733 pci_read_config_dword(dev,
734 pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
735 pci_read_config_dword(dev,
736 pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
737 pci_read_config_dword(dev,
738 pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
739 pci_read_config_dword(dev,
740 pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
741 }
742 }
743
744 return AER_SUCCESS;
745}
746
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800747static inline void aer_process_err_devices(struct pcie_device *p_device,
748 struct aer_err_info *e_info)
749{
750 int i;
751
752 if (!e_info->dev[0]) {
753 dev_printk(KERN_DEBUG, &p_device->port->dev,
754 "can't find device of ID%04x\n",
755 e_info->id);
756 }
757
758 for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
759 if (get_device_error_info(e_info->dev[i], e_info) ==
760 AER_SUCCESS) {
761 aer_print_error(e_info->dev[i], e_info);
762 handle_error_source(p_device,
763 e_info->dev[i],
764 e_info);
765 }
766 }
767}
768
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800769/**
770 * aer_isr_one_error - consume an error detected by root port
771 * @p_device: pointer to error root port service device
772 * @e_src: pointer to an error source
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800773 */
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800774static void aer_isr_one_error(struct pcie_device *p_device,
775 struct aer_err_source *e_src)
776{
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800777 struct aer_err_info *e_info;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800778 int i;
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800779
780 /* struct aer_err_info might be big, so we allocate it with slab */
781 e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL);
782 if (e_info == NULL) {
783 dev_printk(KERN_DEBUG, &p_device->port->dev,
784 "Can't allocate mem when processing AER errors\n");
785 return;
786 }
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800787
788 /*
789 * There is a possibility that both correctable error and
790 * uncorrectable error being logged. Report correctable error first.
791 */
792 for (i = 1; i & ROOT_ERR_STATUS_MASKS ; i <<= 2) {
793 if (i > 4)
794 break;
795 if (!(e_src->status & i))
796 continue;
797
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800798 memset(e_info, 0, sizeof(struct aer_err_info));
799
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800800 /* Init comprehensive error information */
801 if (i & PCI_ERR_ROOT_COR_RCV) {
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800802 e_info->id = ERR_COR_ID(e_src->id);
803 e_info->severity = AER_CORRECTABLE;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800804 } else {
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800805 e_info->id = ERR_UNCOR_ID(e_src->id);
806 e_info->severity = ((e_src->status >> 6) & 1);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800807 }
808 if (e_src->status &
809 (PCI_ERR_ROOT_MULTI_COR_RCV |
810 PCI_ERR_ROOT_MULTI_UNCOR_RCV))
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800811 e_info->flags |= AER_MULTI_ERROR_VALID_FLAG;
812
813 find_source_device(p_device->port, e_info);
Zhang, Yanmin3d5505c2009-06-16 13:35:16 +0800814 aer_process_err_devices(p_device, e_info);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800815 }
Zhang, Yanmin28eb27c2009-06-16 13:35:11 +0800816
817 kfree(e_info);
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800818}
819
820/**
821 * aer_isr - consume errors detected by root port
David Howells65f27f32006-11-22 14:55:48 +0000822 * @work: definition of this work item
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800823 *
824 * Invoked, as DPC, when root port records new detected error
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800825 */
David Howells65f27f32006-11-22 14:55:48 +0000826void aer_isr(struct work_struct *work)
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800827{
David Howells65f27f32006-11-22 14:55:48 +0000828 struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
829 struct pcie_device *p_device = rpc->rpd;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800830 struct aer_err_source *e_src;
831
832 mutex_lock(&rpc->rpc_mutex);
833 e_src = get_e_source(rpc);
834 while (e_src) {
835 aer_isr_one_error(p_device, e_src);
836 e_src = get_e_source(rpc);
837 }
838 mutex_unlock(&rpc->rpc_mutex);
839
840 wake_up(&rpc->wait_release);
841}
842
843/**
844 * aer_delete_rootport - disable root port aer and delete service data
845 * @rpc: pointer to a root port device being deleted
846 *
847 * Invoked when AER service unloaded on a specific Root Port
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800848 */
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800849void aer_delete_rootport(struct aer_rpc *rpc)
850{
851 /* Disable root port AER itself */
852 disable_root_aer(rpc);
853
854 kfree(rpc);
855}
856
857/**
858 * aer_init - provide AER initialization
859 * @dev: pointer to AER pcie device
860 *
861 * Invoked when AER service driver is loaded.
Randy Dunlapd885c6b2007-11-28 09:04:23 -0800862 */
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800863int aer_init(struct pcie_device *dev)
864{
Zhang, Yanmin8d29bfb2007-06-06 11:44:16 +0800865 if (aer_osc_setup(dev) && !forceload)
866 return -ENXIO;
Zhang, Yanmin6c2b3742006-07-31 15:21:33 +0800867
868 return AER_SUCCESS;
869}